{"slug": "deepseek-v4-pro", "name": "DeepSeek V4 Pro", "category": "ai-models", "category_name": "AI models", "description": "DeepSeek's larger V4-series large language model, available through its pay-per-token API.", "recommenders": 82, "commenters": 157, "warners": 12, "score": 77.85, "rising": -14.55, "first": "2026-04-23", "latest": "2026-10-29", "why": "HN commenters use it as a very cheap daily driver for coding and agent work, with costs kept low by deep discounts on cache reads.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 86, "neutral": 41, "compare": 33, "warn": 14}, "spark": [0, 0, 0, 0, 0, 4, 19, 37, 16, 7, 2, 1], "listed": true, "loved": ["Low cost", "Cache pricing"], "caveat": "Frequent mistakes", "short_quote": "Don't sleep on Deepseek, V4 Pro was responsible for the biggest leaps and it cost all of $15. It's genuinely great.", "kind": "", "icon": true, "family": "deepseek", "domain": "deepseek.com", "github_repo": null, "official_url": "https://www.deepseek.com", "pricing": "paid", "aliases": ["DeepSeek V4 Pro", "DeepSeek Pro", "DeepSeek-V4-Pro", "DeepSeek V4 Pro 0813", "DeepSeek V4-Pro", "DeepSeek V4-pro", "Deepseek V4 pro"], "verdict": [{"text": "HN commenters use it as a very cheap daily driver for coding and agent work,", "cites": [{"id": 783856689, "item_id": 48991043, "url": "https://news.ycombinator.com/item?id=48991043", "by": "Squeeze2664", "date": "2026-07-21", "version": "4", "stance": "recommend", "quote": "It's also unbelievably cheap - so cheap, in fact, that I have started using it even for trivial tasks.", "excerpt": "I heavily use the Pi harness with DeepSeek-v4-pro directly from the DeepSeek API almost exclusively and I have have barely ever felt the need for something different. I have some OpenRouter credits and API key ready to go if DeepSeek fails me, and that almost never happens. It's also unbelievably cheap - so cheap, in fact, that I have started using it even for trivial tasks. I use it for coding, code review, managing", "context": "", "use_case": "coding, code review, general questions", "switched_from": "Claude Code", "story": "Ask HN: Claude Code or Codex?"}, {"id": 784014896, "item_id": 49000931, "url": "https://news.ycombinator.com/item?id=49000931", "by": "dools", "date": "2026-07-21", "version": "4", "stance": "recommend", "quote": "Yeah deepseek is my workhorse of choice these days", "excerpt": "Yeah deepseek is my workhorse of choice these days, I mostly use pro because I do a lot of concurrent jobs so speed is less of a concern. When it falls over I change the model mid-chat to GPT5.5 and chuck a couple of tokens over to OpenAI and then once it's correctly found the problem I switch back to deepseek, keeping all the 5.5 analysis in context. I have been a heavy user of k2.5/6 but they seem to have gotten sl", "context": "", "use_case": "concurrent coding jobs", "switched_from": "", "story": "Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA"}, {"id": 787474368, "item_id": 49217148, "url": "https://news.ycombinator.com/item?id=49217148", "by": "dools", "date": "2026-08-07", "version": "4", "stance": "recommend", "quote": "I have been using deepseek v4 pro almost exclusively.", "excerpt": "I have been using deepseek v4 pro almost exclusively. I was using Kimi a lot but it just nose dived. The decline started with the release of 2.7 and accelerated with the release of 3. When I need vision capabilities I use GPT 5.3 codex and if deepseek can\u2019t figure something out after a few goes I switch to GTP 5.5 or 5.6 (I\u2019ve been giving Terra first bite recently and it does pretty well, and have used Sol a couple o", "context": "", "use_case": "coding with agents", "switched_from": "Kimi", "story": "DeepSeek V4 Flash 0731"}, {"id": 783005152, "item_id": 48937822, "url": "https://news.ycombinator.com/item?id=48937822", "by": "ignoramous", "date": "2026-07-16", "version": "4", "stance": "recommend", "quote": "are genuinely good enough as daily drivers, given the costs", "excerpt": "Possible, but pay-as-you-go Hy3 / DeepSeek v4 Pro / MiMo v2.5 Pro (from respective vendors) are genuinely good enough as daily drivers, given the costs (especially, low prices for input cache, which usually makes up 70%+ of total input for agentic workflows). I put in $10 in DeepSeek & Xiaomi MiMo, and I've barely used $1 each, in a week of coding work. Coding Plans by MiniMax ($20/mo for 1.7b tokens) and Z.ai (~$30/", "context": "", "use_case": "daily coding", "switched_from": "", "story": "Kimi K3 is now live"}, {"id": 781350561, "item_id": 48834410, "url": "https://news.ycombinator.com/item?id=48834410", "by": "wongarsu", "date": "2026-07-08", "version": "4", "stance": "recommend", "quote": "For hobby projects I've completely switched to DeepSeek v4 pro.", "excerpt": "At work I wouldn't want to use anything else. Compared to my salary a Claude subscription (or two) is cheap For hobby projects I've completely switched to DeepSeek v4 pro. I spend less than on a $10 Claude plan and am not subjected to quota limits (when I have time and motivation, the last thing I want is a 5 hour quota running out). And the difference in model performance is fine for those smaller projects, most of ", "context": "", "use_case": "hobby projects", "switched_from": "Claude", "story": "SWE-1.7 Reach Near GPT 5.5 and Opus Intelligence"}]}, {"text": " with costs kept low by deep discounts on cache reads.", "cites": [{"id": 788412400, "item_id": 49275775, "url": "https://news.ycombinator.com/item?id=49275775", "by": "xynelius", "date": "2026-08-12", "version": "4", "stance": "recommend", "quote": "it's actually ~60x cheaper if you take into account the typical cache-read/input/output split", "excerpt": "If that wasn't impressive enough, it's actually ~60x cheaper if you take into account the typical cache-read/input/output split in agentic coding, and the deep discount for cache reads offered by DeepSeek. Opencode has some public data on the typical split [1]: For DeepSeek V4 Pro the typical split is 750 in, 290 out, 82k cached. Cost per request for V4 Pro: $0.000875 per request. Equivalent Opus cost (w/o taking int", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 787160624, "item_id": 49197539, "url": "https://news.ycombinator.com/item?id=49197539", "by": "petercooper", "date": "2026-08-06", "version": "4", "stance": "recommend", "quote": "I'm guessing this is almost entirely about the incredibly low cache read prices.", "excerpt": "I'm guessing this is almost entirely about the incredibly low cache read prices. Few have come close to them, nothing has a bigger effect on (a typical) session price, and with the price of RAM right now, they have to be the biggest pain point for them right now? A 10x increase in cache read would be a significant increase, yet would still keep them cheaper than every other provider of their model (at least based on ", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek planning to significantly raise prices"}, {"id": 797043537, "item_id": 49815221, "url": "https://news.ycombinator.com/item?id=49815221", "by": "abalashov", "date": "2026-09-23", "version": "4", "stance": "recommend", "quote": "very cheap large-scale code analyses and things that require mega context windows", "excerpt": "Data points from my own usage, for whatever they're worth[1]. I'm just an individual contributor and hardly represent enterprise or large orgs: I've been on Kimi, with a little DeepSeek-V4-Pro, GLM 5.2/5.3, and MiMo thrown in, for probably about a year now. It's great here! 1) For DeepSeek, I recommend their Reasonix harness strongly, due to its alignment to DeepSeek's prefix cache. It means mostly (95%+) cache hit i", "context": "individual contributor", "use_case": "large codebase analysis", "switched_from": "", "story": "The current balance of power in open models"}, {"id": 783005152, "item_id": 48937822, "url": "https://news.ycombinator.com/item?id=48937822", "by": "ignoramous", "date": "2026-07-16", "version": "4", "stance": "recommend", "quote": "are genuinely good enough as daily drivers, given the costs", "excerpt": "Possible, but pay-as-you-go Hy3 / DeepSeek v4 Pro / MiMo v2.5 Pro (from respective vendors) are genuinely good enough as daily drivers, given the costs (especially, low prices for input cache, which usually makes up 70%+ of total input for agentic workflows). I put in $10 in DeepSeek & Xiaomi MiMo, and I've barely used $1 each, in a week of coding work. Coding Plans by MiniMax ($20/mo for 1.7b tokens) and Z.ai (~$30/", "context": "", "use_case": "daily coding", "switched_from": "", "story": "Kimi K3 is now live"}]}, {"text": " The most common caveat is that it makes mistakes more often than top models,", "cites": [{"id": 787518448, "item_id": 49219903, "url": "https://news.ycombinator.com/item?id=49219903", "by": "KronisLV", "date": "2026-08-08", "version": "4", "stance": "warn", "quote": "the preview version of the model kept regularly messing things up", "excerpt": "> The Chinese models are not good enough for anything other than pair programming, which is just a very last-gen way of using agents. This matches my experience with DeepSeek V4 Pro at Max reasoning, the preview version of the model kept regularly messing things up. About 30-60% of additional time to fix the output was needed. On similar tasks, GLM 5.2 at Max reasoning screwed up maybe 20-30% of the time, while it st", "context": "", "use_case": "Coding agent", "switched_from": "", "story": "DeepSeek V4 Flash 0731"}, {"id": 783050022, "item_id": 48940626, "url": "https://news.ycombinator.com/item?id=48940626", "by": "KronisLV", "date": "2026-07-16", "version": "4", "stance": "warn", "quote": "DeepSeek V4 Pro is really affordable per-token but regularly kept making mistakes in the tasks I gave it.", "excerpt": "> I've yet to find a way to run it on a monthly sub reliably for cheaper than Codex. Matches my experience, I got their Pro subscription and while I enjoyed the model itself a lot and while their ZCode harness is also pretty nice, it gave me less tokens for similar amounts of money that Anthropic would give me on a subscription: https://blog.kronis.dev/blog/z-ai-s-glm-5-2-is-a-great-model-but-is-it-good-value/ I'm ye", "context": "", "use_case": "", "switched_from": "", "story": "Kimi K3 is now live"}, {"id": 788539920, "item_id": 49283745, "url": "https://news.ycombinator.com/item?id=49283745", "by": "arj", "date": "2026-08-13", "version": "4", "stance": "warn", "quote": "Pro feels really slow compared.", "excerpt": "Been testing this on hobby project https://github.com/arj03/seedkernel/. Latest flash was a big step up. Pro feels really slow compared. Claude opus is still better day this level.", "context": "", "use_case": "hobby project coding", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}]}, {"text": " and some users switch to a pricier model when it gets stuck.", "cites": [{"id": 784014896, "item_id": 49000931, "url": "https://news.ycombinator.com/item?id=49000931", "by": "dools", "date": "2026-07-21", "version": "4", "stance": "recommend", "quote": "Yeah deepseek is my workhorse of choice these days", "excerpt": "Yeah deepseek is my workhorse of choice these days, I mostly use pro because I do a lot of concurrent jobs so speed is less of a concern. When it falls over I change the model mid-chat to GPT5.5 and chuck a couple of tokens over to OpenAI and then once it's correctly found the problem I switch back to deepseek, keeping all the 5.5 analysis in context. I have been a heavy user of k2.5/6 but they seem to have gotten sl", "context": "", "use_case": "concurrent coding jobs", "switched_from": "", "story": "Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA"}, {"id": 787474368, "item_id": 49217148, "url": "https://news.ycombinator.com/item?id=49217148", "by": "dools", "date": "2026-08-07", "version": "4", "stance": "recommend", "quote": "I have been using deepseek v4 pro almost exclusively.", "excerpt": "I have been using deepseek v4 pro almost exclusively. I was using Kimi a lot but it just nose dived. The decline started with the release of 2.7 and accelerated with the release of 3. When I need vision capabilities I use GPT 5.3 codex and if deepseek can\u2019t figure something out after a few goes I switch to GTP 5.5 or 5.6 (I\u2019ve been giving Terra first bite recently and it does pretty well, and have used Sol a couple o", "context": "", "use_case": "coding with agents", "switched_from": "Kimi", "story": "DeepSeek V4 Flash 0731"}]}], "praise": [{"label": "Low cost", "summary": "Commenters repeatedly call it cheap enough to use for nearly everything, including trivial tasks.", "quotes": [{"id": 783856689, "item_id": 48991043, "url": "https://news.ycombinator.com/item?id=48991043", "by": "Squeeze2664", "date": "2026-07-21", "version": "4", "stance": "recommend", "quote": "It's also unbelievably cheap - so cheap, in fact, that I have started using it even for trivial tasks.", "excerpt": "I heavily use the Pi harness with DeepSeek-v4-pro directly from the DeepSeek API almost exclusively and I have have barely ever felt the need for something different. I have some OpenRouter credits and API key ready to go if DeepSeek fails me, and that almost never happens. It's also unbelievably cheap - so cheap, in fact, that I have started using it even for trivial tasks. I use it for coding, code review, managing", "context": "", "use_case": "coding, code review, general questions", "switched_from": "Claude Code", "story": "Ask HN: Claude Code or Codex?"}, {"id": 781350561, "item_id": 48834410, "url": "https://news.ycombinator.com/item?id=48834410", "by": "wongarsu", "date": "2026-07-08", "version": "4", "stance": "recommend", "quote": "For hobby projects I've completely switched to DeepSeek v4 pro.", "excerpt": "At work I wouldn't want to use anything else. Compared to my salary a Claude subscription (or two) is cheap For hobby projects I've completely switched to DeepSeek v4 pro. I spend less than on a $10 Claude plan and am not subjected to quota limits (when I have time and motivation, the last thing I want is a 5 hour quota running out). And the difference in model performance is fine for those smaller projects, most of ", "context": "", "use_case": "hobby projects", "switched_from": "Claude", "story": "SWE-1.7 Reach Near GPT 5.5 and Opus Intelligence"}, {"id": 783005152, "item_id": 48937822, "url": "https://news.ycombinator.com/item?id=48937822", "by": "ignoramous", "date": "2026-07-16", "version": "4", "stance": "recommend", "quote": "are genuinely good enough as daily drivers, given the costs", "excerpt": "Possible, but pay-as-you-go Hy3 / DeepSeek v4 Pro / MiMo v2.5 Pro (from respective vendors) are genuinely good enough as daily drivers, given the costs (especially, low prices for input cache, which usually makes up 70%+ of total input for agentic workflows). I put in $10 in DeepSeek & Xiaomi MiMo, and I've barely used $1 each, in a week of coding work. Coding Plans by MiniMax ($20/mo for 1.7b tokens) and Z.ai (~$30/", "context": "", "use_case": "daily coding", "switched_from": "", "story": "Kimi K3 is now live"}]}, {"label": "Cache pricing", "summary": "Low cache-read prices make long agentic sessions and big-context code analysis very cheap.", "quotes": [{"id": 788412400, "item_id": 49275775, "url": "https://news.ycombinator.com/item?id=49275775", "by": "xynelius", "date": "2026-08-12", "version": "4", "stance": "recommend", "quote": "it's actually ~60x cheaper if you take into account the typical cache-read/input/output split", "excerpt": "If that wasn't impressive enough, it's actually ~60x cheaper if you take into account the typical cache-read/input/output split in agentic coding, and the deep discount for cache reads offered by DeepSeek. Opencode has some public data on the typical split [1]: For DeepSeek V4 Pro the typical split is 750 in, 290 out, 82k cached. Cost per request for V4 Pro: $0.000875 per request. Equivalent Opus cost (w/o taking int", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 787160624, "item_id": 49197539, "url": "https://news.ycombinator.com/item?id=49197539", "by": "petercooper", "date": "2026-08-06", "version": "4", "stance": "recommend", "quote": "I'm guessing this is almost entirely about the incredibly low cache read prices.", "excerpt": "I'm guessing this is almost entirely about the incredibly low cache read prices. Few have come close to them, nothing has a bigger effect on (a typical) session price, and with the price of RAM right now, they have to be the biggest pain point for them right now? A 10x increase in cache read would be a significant increase, yet would still keep them cheaper than every other provider of their model (at least based on ", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek planning to significantly raise prices"}, {"id": 797043537, "item_id": 49815221, "url": "https://news.ycombinator.com/item?id=49815221", "by": "abalashov", "date": "2026-09-23", "version": "4", "stance": "recommend", "quote": "very cheap large-scale code analyses and things that require mega context windows", "excerpt": "Data points from my own usage, for whatever they're worth[1]. I'm just an individual contributor and hardly represent enterprise or large orgs: I've been on Kimi, with a little DeepSeek-V4-Pro, GLM 5.2/5.3, and MiMo thrown in, for probably about a year now. It's great here! 1) For DeepSeek, I recommend their Reasonix harness strongly, due to its alignment to DeepSeek's prefix cache. It means mostly (95%+) cache hit i", "context": "individual contributor", "use_case": "large codebase analysis", "switched_from": "", "story": "The current balance of power in open models"}]}, {"label": "Daily workhorse", "summary": "Several users run it almost exclusively as their main model for coding and general work.", "quotes": [{"id": 787474368, "item_id": 49217148, "url": "https://news.ycombinator.com/item?id=49217148", "by": "dools", "date": "2026-08-07", "version": "4", "stance": "recommend", "quote": "I have been using deepseek v4 pro almost exclusively.", "excerpt": "I have been using deepseek v4 pro almost exclusively. I was using Kimi a lot but it just nose dived. The decline started with the release of 2.7 and accelerated with the release of 3. When I need vision capabilities I use GPT 5.3 codex and if deepseek can\u2019t figure something out after a few goes I switch to GTP 5.5 or 5.6 (I\u2019ve been giving Terra first bite recently and it does pretty well, and have used Sol a couple o", "context": "", "use_case": "coding with agents", "switched_from": "Kimi", "story": "DeepSeek V4 Flash 0731"}, {"id": 784014896, "item_id": 49000931, "url": "https://news.ycombinator.com/item?id=49000931", "by": "dools", "date": "2026-07-21", "version": "4", "stance": "recommend", "quote": "Yeah deepseek is my workhorse of choice these days", "excerpt": "Yeah deepseek is my workhorse of choice these days, I mostly use pro because I do a lot of concurrent jobs so speed is less of a concern. When it falls over I change the model mid-chat to GPT5.5 and chuck a couple of tokens over to OpenAI and then once it's correctly found the problem I switch back to deepseek, keeping all the 5.5 analysis in context. I have been a heavy user of k2.5/6 but they seem to have gotten sl", "context": "", "use_case": "concurrent coding jobs", "switched_from": "", "story": "Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA"}, {"id": 790188784, "item_id": 49386799, "url": "https://news.ycombinator.com/item?id=49386799", "by": "try-working", "date": "2026-08-21", "version": "4", "stance": "recommend", "quote": "I main V4 Pro at work now, and at home I route between Pro and Flash based on task.", "excerpt": "I main V4 Pro at work now, and at home I route between Pro and Flash based on task. Switched to Opus 4.6 for some tasks at work because I needed image input - horrible. So nice to get image input with DS.", "context": "", "use_case": "daily work coding", "switched_from": "", "story": "DeepSeek-v4-flash-vision-exp"}]}, {"label": "Pairs with Flash", "summary": "Users route harder planning work to Pro and cheaper detail work to V4 Flash.", "quotes": [{"id": 788411680, "item_id": 49275730, "url": "https://news.ycombinator.com/item?id=49275730", "by": "surgical_fire", "date": "2026-08-12", "version": "4", "stance": "recommend", "quote": "pro plans, flash implements.", "excerpt": "I use a plan -> implement wotkflow for this reason. pro plans, flash implements. I am super happy with how flash behaves like that.", "context": "", "use_case": "planning", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 790188784, "item_id": 49386799, "url": "https://news.ycombinator.com/item?id=49386799", "by": "try-working", "date": "2026-08-21", "version": "4", "stance": "recommend", "quote": "I main V4 Pro at work now, and at home I route between Pro and Flash based on task.", "excerpt": "I main V4 Pro at work now, and at home I route between Pro and Flash based on task. Switched to Opus 4.6 for some tasks at work because I needed image input - horrible. So nice to get image input with DS.", "context": "", "use_case": "daily work coding", "switched_from": "", "story": "DeepSeek-v4-flash-vision-exp"}, {"id": 781038833, "item_id": 48814927, "url": "https://news.ycombinator.com/item?id=48814927", "by": "miroljub", "date": "2026-07-07", "version": "4", "stance": "recommend", "quote": "an occasional usage of DeepSeek V4 Pro", "excerpt": "If you use DeepSeek v4 Flash as a daily driver, with an occasional usage of DeepSeek V4 Pro and Glam 5.2 when necessary, the monthly quota practically never runs out.", "context": "", "use_case": "harder tasks", "switched_from": "", "story": "GLM 5.2 and the coming AI margin collapse"}]}, {"label": "Follows instructions", "summary": "Some use it to implement specs written elsewhere because it does what it is told.", "quotes": [{"id": 785488994, "item_id": 49093062, "url": "https://news.ycombinator.com/item?id=49093062", "by": "wilj", "date": "2026-07-28", "version": "4", "stance": "recommend", "quote": "have Deepseek v4 Pro implement because it just seems to actually do what it's told, given enough tries", "excerpt": "Similar here. I use Claude Code and Codex for developing specs, but then have Deepseek v4 Pro implement because it just seems to actually do what it's told, given enough tries. And Reasonix + Deepseek is mostly cache hits when I let it grind on it for a long time, and I can have a huge fleet of agents with very specific focus for code review and rule enforcement mostly sharing cache.", "context": "", "use_case": "implementing specs", "switched_from": "", "story": "Boris Cherny says \"delete your Claude.md every 6 months\""}, {"id": 783418515, "item_id": 48963657, "url": "https://news.ycombinator.com/item?id=48963657", "by": "rubyn00bie", "date": "2026-07-18", "version": null, "stance": "recommend", "quote": "Deepseek is far and away the most cost effective, fast, and quite good for 90% of implementation work", "excerpt": "I think it depends on the problem space, and to a large extent the languages used. I have had the opposite experience with GPT. I have found GPT to be pretty awful at writing Elixir, and dealing with more open ended tasks. Opus is (for the moment) better at writing Elixir, and Fable is much better than either at actually grokking a problem space. I had switched to GPT for about six weeks and eventually stopped using ", "context": "Elixir developer", "use_case": "implementation work", "switched_from": "", "story": "Fable 5 vs. GPT-5.6 Sol on an NP-Hard Problem: Does /goal help?"}, {"id": 785958352, "item_id": 49122397, "url": "https://news.ycombinator.com/item?id=49122397", "by": "alecsm", "date": "2026-07-31", "version": "4", "stance": "recommend", "quote": "makes DeepSeek Pro the best LLM for me. It allows me to work relatively fast at a very low price.", "excerpt": "I've been using DeepSeek Pro for a while and Flash only for certain dumb tasks where I only need the speed of a LLM and not big brains. I find the newest OpenAI and Anthropic models to be way better for big tasks that require many decisions but I don't like that anyway because I lose track of what's being done. Knowing what I want for every prompt makes DeepSeek Pro the best LLM for me. It allows me to work relativel", "context": "", "use_case": "coding", "switched_from": "", "story": "DeepSeek-V4-Flash Update"}]}, {"label": "Writing quality", "summary": "One commenter finds its writing more nuanced and natural.", "quotes": [{"id": 785977297, "item_id": 49123581, "url": "https://news.ycombinator.com/item?id=49123581", "by": "mcbuilder", "date": "2026-07-31", "version": "4", "stance": "recommend", "quote": "I love deepseek's (Pro) writing.", "excerpt": "I love deepseek's (Pro) writing. I feel it's more nuanced and natural.", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek-V4-Flash Update"}]}], "caveats": [{"label": "Frequent mistakes", "summary": "Some users report regular errors that took extra time to fix.", "quotes": [{"id": 787518448, "item_id": 49219903, "url": "https://news.ycombinator.com/item?id=49219903", "by": "KronisLV", "date": "2026-08-08", "version": "4", "stance": "warn", "quote": "the preview version of the model kept regularly messing things up", "excerpt": "> The Chinese models are not good enough for anything other than pair programming, which is just a very last-gen way of using agents. This matches my experience with DeepSeek V4 Pro at Max reasoning, the preview version of the model kept regularly messing things up. About 30-60% of additional time to fix the output was needed. On similar tasks, GLM 5.2 at Max reasoning screwed up maybe 20-30% of the time, while it st", "context": "", "use_case": "Coding agent", "switched_from": "", "story": "DeepSeek V4 Flash 0731"}, {"id": 783050022, "item_id": 48940626, "url": "https://news.ycombinator.com/item?id=48940626", "by": "KronisLV", "date": "2026-07-16", "version": "4", "stance": "warn", "quote": "DeepSeek V4 Pro is really affordable per-token but regularly kept making mistakes in the tasks I gave it.", "excerpt": "> I've yet to find a way to run it on a monthly sub reliably for cheaper than Codex. Matches my experience, I got their Pro subscription and while I enjoyed the model itself a lot and while their ZCode harness is also pretty nice, it gave me less tokens for similar amounts of money that Anthropic would give me on a subscription: https://blog.kronis.dev/blog/z-ai-s-glm-5-2-is-a-great-model-but-is-it-good-value/ I'm ye", "context": "", "use_case": "", "switched_from": "", "story": "Kimi K3 is now live"}]}, {"label": "Fallback needed", "summary": "When it gets stuck, users switch to GPT models to find the problem.", "quotes": [{"id": 784014896, "item_id": 49000931, "url": "https://news.ycombinator.com/item?id=49000931", "by": "dools", "date": "2026-07-21", "version": "4", "stance": "recommend", "quote": "Yeah deepseek is my workhorse of choice these days", "excerpt": "Yeah deepseek is my workhorse of choice these days, I mostly use pro because I do a lot of concurrent jobs so speed is less of a concern. When it falls over I change the model mid-chat to GPT5.5 and chuck a couple of tokens over to OpenAI and then once it's correctly found the problem I switch back to deepseek, keeping all the 5.5 analysis in context. I have been a heavy user of k2.5/6 but they seem to have gotten sl", "context": "", "use_case": "concurrent coding jobs", "switched_from": "", "story": "Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA"}, {"id": 787474368, "item_id": 49217148, "url": "https://news.ycombinator.com/item?id=49217148", "by": "dools", "date": "2026-08-07", "version": "4", "stance": "recommend", "quote": "I have been using deepseek v4 pro almost exclusively.", "excerpt": "I have been using deepseek v4 pro almost exclusively. I was using Kimi a lot but it just nose dived. The decline started with the release of 2.7 and accelerated with the release of 3. When I need vision capabilities I use GPT 5.3 codex and if deepseek can\u2019t figure something out after a few goes I switch to GTP 5.5 or 5.6 (I\u2019ve been giving Terra first bite recently and it does pretty well, and have used Sol a couple o", "context": "", "use_case": "coding with agents", "switched_from": "Kimi", "story": "DeepSeek V4 Flash 0731"}]}, {"label": "Big tasks", "summary": "Newer OpenAI, Anthropic and Claude Opus models are seen as better for complex tasks.", "quotes": [{"id": 785958352, "item_id": 49122397, "url": "https://news.ycombinator.com/item?id=49122397", "by": "alecsm", "date": "2026-07-31", "version": "4", "stance": "recommend", "quote": "makes DeepSeek Pro the best LLM for me. It allows me to work relatively fast at a very low price.", "excerpt": "I've been using DeepSeek Pro for a while and Flash only for certain dumb tasks where I only need the speed of a LLM and not big brains. I find the newest OpenAI and Anthropic models to be way better for big tasks that require many decisions but I don't like that anyway because I lose track of what's being done. Knowing what I want for every prompt makes DeepSeek Pro the best LLM for me. It allows me to work relativel", "context": "", "use_case": "coding", "switched_from": "", "story": "DeepSeek-V4-Flash Update"}, {"id": 788539920, "item_id": 49283745, "url": "https://news.ycombinator.com/item?id=49283745", "by": "arj", "date": "2026-08-13", "version": "4", "stance": "warn", "quote": "Pro feels really slow compared.", "excerpt": "Been testing this on hobby project https://github.com/arj03/seedkernel/. Latest flash was a big step up. Pro feels really slow compared. Claude opus is still better day this level.", "context": "", "use_case": "hobby project coding", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}]}, {"label": "Slow", "summary": "One tester found Pro really slow compared with Flash.", "quotes": [{"id": 788539920, "item_id": 49283745, "url": "https://news.ycombinator.com/item?id=49283745", "by": "arj", "date": "2026-08-13", "version": "4", "stance": "warn", "quote": "Pro feels really slow compared.", "excerpt": "Been testing this on hobby project https://github.com/arj03/seedkernel/. Latest flash was a big step up. Pro feels really slow compared. Claude opus is still better day this level.", "context": "", "use_case": "hobby project coding", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 784014896, "item_id": 49000931, "url": "https://news.ycombinator.com/item?id=49000931", "by": "dools", "date": "2026-07-21", "version": "4", "stance": "recommend", "quote": "Yeah deepseek is my workhorse of choice these days", "excerpt": "Yeah deepseek is my workhorse of choice these days, I mostly use pro because I do a lot of concurrent jobs so speed is less of a concern. When it falls over I change the model mid-chat to GPT5.5 and chuck a couple of tokens over to OpenAI and then once it's correctly found the problem I switch back to deepseek, keeping all the 5.5 analysis in context. I have been a heavy user of k2.5/6 but they seem to have gotten sl", "context": "", "use_case": "concurrent coding jobs", "switched_from": "", "story": "Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA"}]}, {"label": "API vs subscription", "summary": "Paying API prices can cost more than a flat subscription to another model.", "quotes": [{"id": 790968384, "item_id": 49435524, "url": "https://news.ycombinator.com/item?id=49435524", "by": "phiresky", "date": "2026-08-25", "version": "4", "stance": "warn", "quote": "I tried for example Deepseek V4 Pro for a bit (via API pricing), but it's _more expensive_ than Sol via subscription", "excerpt": "I think you're underestimating the \"more expensive\". It's not a 30% discount or similar. API is at least 10 times more expensive (own experience), ~40x as expensive if [1] is accurate. At that price difference, lots of things that are viable to do in whatever harness the subscription allows are simply impractical via API, you'd be spending a whole developer's salary just for the benefit of using a different harness. ", "context": "", "use_case": "", "switched_from": "", "story": "OpenAI restores 5-hour Codex and Work limits for ChatGPT Plus users"}, {"id": 783028384, "item_id": 48939274, "url": "https://news.ycombinator.com/item?id=48939274", "by": "cg5280", "date": "2026-07-16", "version": "4", "stance": "recommend", "quote": "I have been using Deepseek V4 Pro for personal projects and it has been great.", "excerpt": "I have been using Deepseek V4 Pro for personal projects and it has been great. I think the $20/mo GPT plan is still the strongest value, but only because you don\u2019t have to pay API prices for tokens.", "context": "", "use_case": "personal projects", "switched_from": "", "story": "Kimi K3 is now live"}]}, {"label": "Data and drift", "summary": "The Reasonix harness sends data to China, and very large contexts bring some attentional drift.", "quotes": [{"id": 796953681, "item_id": 49809605, "url": "https://news.ycombinator.com/item?id=49809605", "by": "abalashov", "date": "2026-09-22", "version": "4", "stance": "recommend", "quote": "very cheap large-scale code analyses and things that require mega context windows", "excerpt": "Welcome to the dark side. I've been on Kimi, with a little DeepSeek-V4-Pro, GLM 5.2/5.3, and MiMo thrown in, for probably about a year now. It's great here! For DeepSeek, I recommend their Reasonix harness strongly, due to its alignment to DeepSeek's prefix cache. It means mostly (95%+) cache hit input tokens, so very cheap large-scale code analyses and things that require mega context windows (at the cost of some at", "context": "", "use_case": "large-scale code analysis", "switched_from": "", "story": "MiMo-v2.6-Pro: Intelligence, Performance and Price Analysis"}, {"id": 797043537, "item_id": 49815221, "url": "https://news.ycombinator.com/item?id=49815221", "by": "abalashov", "date": "2026-09-23", "version": "4", "stance": "recommend", "quote": "very cheap large-scale code analyses and things that require mega context windows", "excerpt": "Data points from my own usage, for whatever they're worth[1]. I'm just an individual contributor and hardly represent enterprise or large orgs: I've been on Kimi, with a little DeepSeek-V4-Pro, GLM 5.2/5.3, and MiMo thrown in, for probably about a year now. It's great here! 1) For DeepSeek, I recommend their Reasonix harness strongly, due to its alignment to DeepSeek's prefix cache. It means mostly (95%+) cache hit i", "context": "individual contributor", "use_case": "large codebase analysis", "switched_from": "", "story": "The current balance of power in open models"}]}], "who": [{"by": "abalashov", "context": "individual contributor", "url": "https://news.ycombinator.com/item?id=49815221", "date": "2026-09-23"}, {"by": "assafa", "context": "side project developer", "url": "https://news.ycombinator.com/item?id=49762665", "date": "2026-09-19"}, {"by": "solarkraft", "context": "hobby project developer", "url": "https://news.ycombinator.com/item?id=49080474", "date": "2026-07-28"}, {"by": "rubyn00bie", "context": "Elixir developer", "url": "https://news.ycombinator.com/item?id=48963657", "date": "2026-07-18"}, {"by": "epolanski", "context": "does not vibe code at work", "url": "https://news.ycombinator.com/item?id=48584034", "date": "2026-06-18"}, {"by": "mdasen", "context": "heavy cached-input API user", "url": "https://news.ycombinator.com/item?id=48507479", "date": "2026-06-12"}, {"by": "trollbridge", "context": "leads a team of 3 developers, 1 designer, 1 non-technical user", "url": "https://news.ycombinator.com/item?id=48503533", "date": "2026-06-12"}, {"by": "caleblloyd", "context": "enterprise software team", "url": "https://news.ycombinator.com/item?id=48471771", "date": "2026-06-09"}, {"by": "rurban", "context": "mostly C, C++, perl and python developer", "url": "https://news.ycombinator.com/item?id=48423534", "date": "2026-06-06"}, {"by": "malshe", "context": "academic researcher", "url": "https://news.ycombinator.com/item?id=48322949", "date": "2026-05-29"}, {"by": "mchusma", "context": "builds vertical SaaS applications", "url": "https://news.ycombinator.com/item?id=47891197", "date": "2026-04-24"}], "tags": ["for coding agents", "for hobby projects", "for large-context code analysis", "for budget-conscious devs"], "use_cases": ["coding", "ai coding", "personal projects", "code review", "harder tasks", "complex coding tasks"], "compared": [{"slug": "claude-opus", "name": "Claude Opus", "co": 44, "switched_from": 2, "switched_to": 0}, {"slug": "glm", "name": "GLM", "co": 41, "switched_from": 1, "switched_to": 1}, {"slug": "deepseek-v4-flash", "name": "DeepSeek V4 Flash", "co": 43, "switched_from": 0, "switched_to": 1}, {"slug": "kimi", "name": "Kimi", "co": 31, "switched_from": 2, "switched_to": 0}, {"slug": "claude", "name": "Claude", "co": 20, "switched_from": 5, "switched_to": 0}, {"slug": "claude-code", "name": "Claude Code", "co": 23, "switched_from": 2, "switched_to": 0}, {"slug": "gpt-5", "name": "GPT-5", "co": 25, "switched_from": 1, "switched_to": 0}, {"slug": "opencode", "name": "OpenCode", "co": 25, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=48237663", "title": "DeepSeek makes the V4 Pro price discount permanent", "mentions": 13}, {"url": "https://news.ycombinator.com/item?id=49274600", "title": "DeepSeek V4 Pro 0813", "mentions": 11}, {"url": "https://news.ycombinator.com/item?id=48440448", "title": "DeepSeek V4 Pro beats GPT-5.5 Pro on precision", "mentions": 9}, {"url": "https://news.ycombinator.com/item?id=47977026", "title": "DeepSeek V4\u2013almost on the frontier, a fraction of the price", "mentions": 9}, {"url": "https://news.ycombinator.com/item?id=47884971", "title": "DeepSeek v4", "mentions": 9}, {"url": "https://news.ycombinator.com/item?id=48002136", "title": "DeepClaude \u2013 Claude Code agent loop with DeepSeek V4 Pro, 17x cheaper", "mentions": 7}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {}, "2026-04": {"compare": 3, "neutral": 3, "recommend": 4, "warn": 1}, "2026-05": {"recommend": 19, "compare": 9, "warn": 6, "neutral": 18}, "2026-06": {"recommend": 37, "neutral": 10, "warn": 4, "compare": 8}, "2026-07": {"recommend": 16, "compare": 6, "neutral": 4}, "2026-08": {"recommend": 7, "warn": 3, "compare": 5, "neutral": 4}, "2026-09": {"neutral": 1, "compare": 2, "recommend": 2}, "2026-10": {"recommend": 1, "neutral": 1}}, "versions": [], "version_unnamed": 7, "family_ref": {"slug": "deepseek", "name": "DeepSeek"}, "runs_on": null, "lines": [], "quotes": [{"id": 804711744, "item_id": 50294484, "url": "https://news.ycombinator.com/item?id=50294484", "by": "badsectoracula", "date": "2026-10-29", "version": "4", "stance": "neutral", "quote": "all their recent stuff (e.g. [0][1]) are MIT licensed", "excerpt": "This is ancient (2 years ago), all their recent stuff (e.g. [0][1]) are MIT licensed. The only exception i've found from a quick browse of their stuff is [2], which is Apache 2 licensed. Even the update to V3 (which you linked at), V3.2, was MIT licensed. [0] https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017 [1] https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813 [2] https://huggingface.co/deepseek-ai/D", "context": "", "use_case": "", "switched_from": "", "story": "We're releasing the weights of Mistral Large 4"}, {"id": 798768608, "item_id": 49923038, "url": "https://news.ycombinator.com/item?id=49923038", "by": "marknutter", "date": "2026-10-01", "version": "4", "stance": "recommend", "quote": "", "excerpt": "damn, I might have to give it a shot myself", "context": "", "use_case": "", "switched_from": "", "story": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price"}, {"id": 797043537, "item_id": 49815221, "url": "https://news.ycombinator.com/item?id=49815221", "by": "abalashov", "date": "2026-09-23", "version": "4", "stance": "recommend", "quote": "very cheap large-scale code analyses and things that require mega context windows", "excerpt": "Data points from my own usage, for whatever they're worth[1]. I'm just an individual contributor and hardly represent enterprise or large orgs: I've been on Kimi, with a little DeepSeek-V4-Pro, GLM 5.2/5.3, and MiMo thrown in, for probably about a year now. It's great here! 1) For DeepSeek, I recommend their Reasonix harness strongly, due to its alignment to DeepSeek's prefix cache. It means mostly (95%+) cache hit i", "context": "individual contributor", "use_case": "large codebase analysis", "switched_from": "", "story": "The current balance of power in open models"}, {"id": 796953681, "item_id": 49809605, "url": "https://news.ycombinator.com/item?id=49809605", "by": "abalashov", "date": "2026-09-22", "version": "4", "stance": "recommend", "quote": "very cheap large-scale code analyses and things that require mega context windows", "excerpt": "Welcome to the dark side. I've been on Kimi, with a little DeepSeek-V4-Pro, GLM 5.2/5.3, and MiMo thrown in, for probably about a year now. It's great here! For DeepSeek, I recommend their Reasonix harness strongly, due to its alignment to DeepSeek's prefix cache. It means mostly (95%+) cache hit input tokens, so very cheap large-scale code analyses and things that require mega context windows (at the cost of some at", "context": "", "use_case": "large-scale code analysis", "switched_from": "", "story": "MiMo-v2.6-Pro: Intelligence, Performance and Price Analysis"}, {"id": 796202641, "item_id": 49762665, "url": "https://news.ycombinator.com/item?id=49762665", "by": "assafa", "date": "2026-09-19", "version": "4", "stance": "recommend", "quote": "I switched to DeepSeek V4 Flash and Pro for a while and I enjoyed that it was cheap enough to get work done.", "excerpt": "DeepSeek V4 Flash. I used Claude before and loved it, no complaint other than it cost more than I\u2019m eager to pay. It's for a side project, so no employer that foots the bill. Anyway, I switched to DeepSeek V4 Flash and Pro for a while and I enjoyed that it was cheap enough to get work done. I also tried GLM 5.3/5.3-Flash for a while. The logic behind it: use the more powerful model for deep thinking work, flip to che", "context": "side project developer", "use_case": "deep thinking work", "switched_from": "Claude", "story": "Ask HN: What default model do you use and why?"}, {"id": 795254549, "item_id": 49703409, "url": "https://news.ycombinator.com/item?id=49703409", "by": "criley2", "date": "2026-09-14", "version": "4", "stance": "compare", "quote": "My experience is that open weights models find between 1/4 to 1/2 of what Fable/Opus stack can find", "excerpt": "I have been writing an internal code review tool that is a bit maximalist. I created subagents for many internal domains and technologies we manage, with prompts focused on best practices, common problems, owasp guidelines, etc), a separate tier of wider band subagents (design, rollout, security/privacy), and a final agent at the top orchestrating and combining. I also use adversarial validator passes against all fin", "context": "", "use_case": "code review", "switched_from": "", "story": "GPT-5.6 Luna vs. GPT-6 Astra: Is a $1.20 Model Good Enough for Code Review?"}, {"id": 794032385, "item_id": 49627024, "url": "https://news.ycombinator.com/item?id=49627024", "by": "KyleTheDev", "date": "2026-09-09", "version": "4", "stance": "compare", "quote": "and much cheaper than the current V4 Pro model", "excerpt": "Direct 1:1:1 comparison, for V4.1 Flash - V4 Pro 0813 - V4 Flash 0731 Input cache hits (per 1m tokens) - $0.003 Vs. $0.022 Vs. $0.007 Input cache miss (per 1m tokens) - $0.15 Vs. $0.66 Vs. $0.22 Output (per 1m tokens) - $0.6 Vs. $1.98 Vs. $0.66 This is taken from https://api-docs.deepseek.com/quick_start/pricing, and it's comparing only off-peak hours pricing. It looks like V4.1 Flash is cheaper than the current 0731", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek launching v4.1 flash cheaper and more capable than v4 pro"}, {"id": 792866801, "item_id": 49554175, "url": "https://news.ycombinator.com/item?id=49554175", "by": "umajho", "date": "2026-09-03", "version": "4", "stance": "neutral", "quote": "all other larger models I tested (DeepSeek V4 Pro, Qwen3.8 Max, GLM 5.3, Kimi K3) drew the pelican riding a bicycle", "excerpt": "I actually did something similar last month. I just asked the LLMs: > Do Simon Willison's classic pelican test. Recall the specification first and then draw it. Qwen3.8-27b didn't know that the test is about riding a bicycle, while all other larger models I tested (DeepSeek V4 Pro, Qwen3.8 Max, GLM 5.3, Kimi K3) drew the pelican riding a bicycle. logs: https://gist.github.com/umajho/c0e20d245d721d7c472a32d317640cb4 r", "context": "", "use_case": "", "switched_from": "", "story": "Muse Spark 1.3"}, {"id": 790968384, "item_id": 49435524, "url": "https://news.ycombinator.com/item?id=49435524", "by": "phiresky", "date": "2026-08-25", "version": "4", "stance": "warn", "quote": "I tried for example Deepseek V4 Pro for a bit (via API pricing), but it's _more expensive_ than Sol via subscription", "excerpt": "I think you're underestimating the \"more expensive\". It's not a 30% discount or similar. API is at least 10 times more expensive (own experience), ~40x as expensive if [1] is accurate. At that price difference, lots of things that are viable to do in whatever harness the subscription allows are simply impractical via API, you'd be spending a whole developer's salary just for the benefit of using a different harness. ", "context": "", "use_case": "", "switched_from": "", "story": "OpenAI restores 5-hour Codex and Work limits for ChatGPT Plus users"}, {"id": 790188784, "item_id": 49386799, "url": "https://news.ycombinator.com/item?id=49386799", "by": "try-working", "date": "2026-08-21", "version": "4", "stance": "recommend", "quote": "I main V4 Pro at work now, and at home I route between Pro and Flash based on task.", "excerpt": "I main V4 Pro at work now, and at home I route between Pro and Flash based on task. Switched to Opus 4.6 for some tasks at work because I needed image input - horrible. So nice to get image input with DS.", "context": "", "use_case": "daily work coding", "switched_from": "", "story": "DeepSeek-v4-flash-vision-exp"}, {"id": 788667873, "item_id": 49291742, "url": "https://news.ycombinator.com/item?id=49291742", "by": "YetAnotherNick", "date": "2026-08-13", "version": "4", "stance": "compare", "quote": "Deepseek v4 pro uses 2.4GB for the same context length", "excerpt": "LLama 3 405B had the most unoptimized kv cache usage by far. Deepseek v4 pro uses 2.4GB for the same context length[1]. [1]: https://vllm.ai/blog/2026-04-24-deepseek-v4", "context": "", "use_case": "", "switched_from": "", "story": "Accelerating GPT-5.6 Sol Ultrafast"}, {"id": 788628896, "item_id": 49289306, "url": "https://news.ycombinator.com/item?id=49289306", "by": "anigbrowl", "date": "2026-08-13", "version": "4", "stance": "neutral", "quote": "Briefly Pro is $2/1m output in off-peak periods, $4 in peak.", "excerpt": "It has been updated now, and will take effect next week: https://api-docs.deepseek.com/quick_start/pricing/ Briefly Pro is $2/1m output in off-peak periods, $4 in peak. Flash is $0.66/$1.32. Input tokens are still much cheaper. I don't mind these prices but I find the need to check against two different time brackets of unequal length an annoying distraction. I guess I need to make some little background app or plugi", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 788612816, "item_id": 49288301, "url": "https://news.ycombinator.com/item?id=49288301", "by": "dhx", "date": "2026-08-13", "version": "4", "stance": "compare", "quote": "DeepSeek v4 Pro 0813 is less accurate with a 35% false positive rate versus GPT-5.6-Sol's 15% false positive rate.", "excerpt": "Now that people have played with the model for a bit, there is one reported real world win for DeepSeek v4 Pro 0813 that is perhaps quite consequential given the drama in the US about Mythos. In a benchmark, DeepSeek v4 Pro 0813 found 87.5% of selected real world software vulnerabilities publicly reported and with CVEs assigned, which is above runner ups Opus 5 and Qwen 3.8 which both found only 81.3%. However there ", "context": "", "use_case": "", "switched_from": "", "story": "Qwen/Qwen3.8-2.4T-A95B"}, {"id": 788539920, "item_id": 49283745, "url": "https://news.ycombinator.com/item?id=49283745", "by": "arj", "date": "2026-08-13", "version": "4", "stance": "warn", "quote": "Pro feels really slow compared.", "excerpt": "Been testing this on hobby project https://github.com/arj03/seedkernel/. Latest flash was a big step up. Pro feels really slow compared. Claude opus is still better day this level.", "context": "", "use_case": "hobby project coding", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 788527312, "item_id": 49282957, "url": "https://news.ycombinator.com/item?id=49282957", "by": "lionkor", "date": "2026-08-13", "version": "4", "stance": "neutral", "quote": "For tasks like pondering on something, reviewing code, etc. I use Pro", "excerpt": "The idea is, I believe, that the Pro model is a larger model (more parameters, or less quantization) in general. What implication that has, I couldn't tell you. For tasks like pondering on something, reviewing code, etc. I use Pro, just because it feels like the right model for that.", "context": "", "use_case": "reviewing code, reasoning", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 788520880, "item_id": 49282555, "url": "https://news.ycombinator.com/item?id=49282555", "by": "WASDx", "date": "2026-08-12", "version": "4", "stance": "compare", "quote": "Why would I use pro over flash now?", "excerpt": "This was a disappointed to me. Why would I use pro over flash now? Is there some area where the difference is significant?", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 788508880, "item_id": 49281805, "url": "https://news.ycombinator.com/item?id=49281805", "by": "wmedrano", "date": "2026-08-12", "version": "4", "stance": "compare", "quote": "Seems a bit weaker than Opus 4.8 but at least 10x cheaper once you account for verbosity.", "excerpt": "Some benchmarks are out. Seems a bit weaker than Opus 4.8 but at least 10x cheaper once you account for verbosity. https://artificialanalysis.ai/?models=claude-opus-5%2Cclaude-opus-4-8%2Cdeepseek-v4-pro%2Cdeepseek-v4-flash%2Cclaude-opus-5-medium%2Cclaude-opus-5-high Wonder how much more they'll squeeze out.", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 788434560, "item_id": 49277160, "url": "https://news.ycombinator.com/item?id=49277160", "by": "taosx", "date": "2026-08-12", "version": "4", "stance": "recommend", "quote": "cheaper to use ds-v4-pro than the luna models from openai", "excerpt": "I created a simulation for coding harnesses based on my own pi sessions. When taking into account all factors, DS-v4-Pro is cheaper than gpt-5.6-luna due to caching. Look at the bill segments difference for cache read cost and uncached cost between deepseek and the other models. At this point is cheaper to use ds-v4-pro than the luna models from openai. ignore the numbers except the classic and keep in mind that clas", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 788412400, "item_id": 49275775, "url": "https://news.ycombinator.com/item?id=49275775", "by": "xynelius", "date": "2026-08-12", "version": "4", "stance": "recommend", "quote": "it's actually ~60x cheaper if you take into account the typical cache-read/input/output split", "excerpt": "If that wasn't impressive enough, it's actually ~60x cheaper if you take into account the typical cache-read/input/output split in agentic coding, and the deep discount for cache reads offered by DeepSeek. Opencode has some public data on the typical split [1]: For DeepSeek V4 Pro the typical split is 750 in, 290 out, 82k cached. Cost per request for V4 Pro: $0.000875 per request. Equivalent Opus cost (w/o taking int", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 788411680, "item_id": 49275730, "url": "https://news.ycombinator.com/item?id=49275730", "by": "surgical_fire", "date": "2026-08-12", "version": "4", "stance": "recommend", "quote": "pro plans, flash implements.", "excerpt": "I use a plan -> implement wotkflow for this reason. pro plans, flash implements. I am super happy with how flash behaves like that.", "context": "", "use_case": "planning", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 788411344, "item_id": 49275709, "url": "https://news.ycombinator.com/item?id=49275709", "by": "parsimo2010", "date": "2026-08-12", "version": "4", "stance": "recommend", "quote": "it still would be a better value to use DS", "excerpt": "The timing looks like they are trying to take the wind out of Qwen's sails by releasing this on the same day that Qwen released the weights of Qwen3.8-max. Or maybe it's coincidence... For comparison I looked at Qwen's claimed benchmarks for Qwen3.8-max (https://qwen.ai/blog?id=qwen3.8). Assuming each published set of benchmarks is believable, it looks like v4 Pro 0813 is better on average but overall performance is ", "context": "", "use_case": "cheap AI model use", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 788409377, "item_id": 49275586, "url": "https://news.ycombinator.com/item?id=49275586", "by": "nolist_policy", "date": "2026-08-12", "version": "4", "stance": "neutral", "quote": "", "excerpt": "DeepSeek V4 Flash is the \"too cheap to meter\" of AI. And you can run the full unquantized model locally for $8000 (2x DGX Spark) at full 1M context and decent speeds: https://github.com/elsung/dgx-spark-deepseek-v4-flash#-long-context--what-to-expect", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 788395504, "item_id": 49274719, "url": "https://news.ycombinator.com/item?id=49274719", "by": "svantana", "date": "2026-08-12", "version": "4", "stance": "neutral", "quote": "just now available on OpenRouter", "excerpt": "And relatedly, just now available on OpenRouter https://openrouter.ai/deepseek/deepseek-v4-pro-0813", "context": "", "use_case": "", "switched_from": "", "story": "Qwen/Qwen3.8-2.4T-A95B"}, {"id": 788393888, "item_id": 49274618, "url": "https://news.ycombinator.com/item?id=49274618", "by": "aabdi", "date": "2026-08-12", "version": "4", "stance": "compare", "quote": "Competitive with opus 4.8 but weaker than sol or fable", "excerpt": "https://api-docs.deepseek.com/quick_start/pricing/ Competitive with opus 4.8 but weaker than sol or fable. About 20x cheaper.", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 787518448, "item_id": 49219903, "url": "https://news.ycombinator.com/item?id=49219903", "by": "KronisLV", "date": "2026-08-08", "version": "4", "stance": "warn", "quote": "the preview version of the model kept regularly messing things up", "excerpt": "> The Chinese models are not good enough for anything other than pair programming, which is just a very last-gen way of using agents. This matches my experience with DeepSeek V4 Pro at Max reasoning, the preview version of the model kept regularly messing things up. About 30-60% of additional time to fix the output was needed. On similar tasks, GLM 5.2 at Max reasoning screwed up maybe 20-30% of the time, while it st", "context": "", "use_case": "Coding agent", "switched_from": "", "story": "DeepSeek V4 Flash 0731"}, {"id": 787474368, "item_id": 49217148, "url": "https://news.ycombinator.com/item?id=49217148", "by": "dools", "date": "2026-08-07", "version": "4", "stance": "recommend", "quote": "I have been using deepseek v4 pro almost exclusively.", "excerpt": "I have been using deepseek v4 pro almost exclusively. I was using Kimi a lot but it just nose dived. The decline started with the release of 2.7 and accelerated with the release of 3. When I need vision capabilities I use GPT 5.3 codex and if deepseek can\u2019t figure something out after a few goes I switch to GTP 5.5 or 5.6 (I\u2019ve been giving Terra first bite recently and it does pretty well, and have used Sol a couple o", "context": "", "use_case": "coding with agents", "switched_from": "Kimi", "story": "DeepSeek V4 Flash 0731"}, {"id": 787160624, "item_id": 49197539, "url": "https://news.ycombinator.com/item?id=49197539", "by": "petercooper", "date": "2026-08-06", "version": "4", "stance": "recommend", "quote": "I'm guessing this is almost entirely about the incredibly low cache read prices.", "excerpt": "I'm guessing this is almost entirely about the incredibly low cache read prices. Few have come close to them, nothing has a bigger effect on (a typical) session price, and with the price of RAM right now, they have to be the biggest pain point for them right now? A 10x increase in cache read would be a significant increase, yet would still keep them cheaper than every other provider of their model (at least based on ", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek planning to significantly raise prices"}, {"id": 786056176, "item_id": 49128511, "url": "https://news.ycombinator.com/item?id=49128511", "by": "fcanesin", "date": "2026-07-31", "version": "4", "stance": "recommend", "quote": "High hopes for V4 Pro", "excerpt": "LOL https://artificialanalysis.ai/models/deepseek-v4-flash#intelligence-comparison-tabs High hopes for V4 Pro", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Flash 0731 Intelligence, Performance and Price Analysis"}, {"id": 785977297, "item_id": 49123581, "url": "https://news.ycombinator.com/item?id=49123581", "by": "mcbuilder", "date": "2026-07-31", "version": "4", "stance": "recommend", "quote": "I love deepseek's (Pro) writing.", "excerpt": "I love deepseek's (Pro) writing. I feel it's more nuanced and natural.", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek-V4-Flash Update"}, {"id": 785958352, "item_id": 49122397, "url": "https://news.ycombinator.com/item?id=49122397", "by": "alecsm", "date": "2026-07-31", "version": "4", "stance": "recommend", "quote": "makes DeepSeek Pro the best LLM for me. It allows me to work relatively fast at a very low price.", "excerpt": "I've been using DeepSeek Pro for a while and Flash only for certain dumb tasks where I only need the speed of a LLM and not big brains. I find the newest OpenAI and Anthropic models to be way better for big tasks that require many decisions but I don't like that anyway because I lose track of what's being done. Knowing what I want for every prompt makes DeepSeek Pro the best LLM for me. It allows me to work relativel", "context": "", "use_case": "coding", "switched_from": "", "story": "DeepSeek-V4-Flash Update"}, {"id": 785942210, "item_id": 49121388, "url": "https://news.ycombinator.com/item?id=49121388", "by": "gpugreg", "date": "2026-07-31", "version": "4", "stance": "neutral", "quote": "For DeepSeek-V4-Pro, the amount is $15.", "excerpt": "To add to this, the $60 only applies to DeepSeek-V4-Flash and a few other models. For DeepSeek-V4-Pro, the amount is $15. https://opencode.ai/docs/go/#usage-limits Previously, OpenCode Go had higher API prices for some models, but now they lowered the API price and simultaneously reduced the allowance.", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek-V4-Flash Update"}, {"id": 785901105, "item_id": 49118819, "url": "https://news.ycombinator.com/item?id=49118819", "by": "jmolinski", "date": "2026-07-30", "version": "4", "stance": "compare", "quote": "cheaper per task than DS", "excerpt": "You are comparing Luna medium to Deepseek Max, when you switch to Luna high it's cheaper per task than DS and has a better intelligence score, which goes higher yet on xhigh and max reasoning levels. https://artificialanalysis.ai/models/comparisons/gpt-5-6-luna-high-vs-deepseek-v4-pro", "context": "", "use_case": "", "switched_from": "", "story": "Advancing the price-performance frontier with GPT\u20115.6"}, {"id": 785824049, "item_id": 49114003, "url": "https://news.ycombinator.com/item?id=49114003", "by": "efficax", "date": "2026-07-30", "version": "4", "stance": "compare", "quote": "here's one measurement that says v4 pro is better than luna", "excerpt": "is it? i don't know how we measure these things, but here's one measurement that says v4 pro is better than luna: https://artificialanalysis.ai/models/comparisons/gpt-5-6-luna-medium-vs-deepseek-v4-pro presumably it's a much bigger model", "context": "", "use_case": "", "switched_from": "", "story": "Advancing the price-performance frontier with GPT\u20115.6"}, {"id": 785488994, "item_id": 49093062, "url": "https://news.ycombinator.com/item?id=49093062", "by": "wilj", "date": "2026-07-28", "version": "4", "stance": "recommend", "quote": "have Deepseek v4 Pro implement because it just seems to actually do what it's told, given enough tries", "excerpt": "Similar here. I use Claude Code and Codex for developing specs, but then have Deepseek v4 Pro implement because it just seems to actually do what it's told, given enough tries. And Reasonix + Deepseek is mostly cache hits when I let it grind on it for a long time, and I can have a huge fleet of agents with very specific focus for code review and rule enforcement mostly sharing cache.", "context": "", "use_case": "implementing specs", "switched_from": "", "story": "Boris Cherny says \"delete your Claude.md every 6 months\""}, {"id": 785312917, "item_id": 49082057, "url": "https://news.ycombinator.com/item?id=49082057", "by": "Frannky", "date": "2026-07-28", "version": null, "stance": "recommend", "quote": "cheap Chinese models (DeepSeek Pro, Flash, MiMo 2.5 Pro for reasoning, and MiniMax M3 for agents)", "excerpt": "I think they're doing something wrong. I have access to better models, but I just use 4.6 and 4.8. Fable and Opus 5 burn tokens too fast for me. I also switched to a cheaper plan to work less and to experiment with OpenCode and open models when I hit the 5h limits. I'm doing this partly because I feel they've already started the enshittification of the old models, and I want to be able to just stop paying for the sub", "context": "", "use_case": "coding", "switched_from": "", "story": "\"Opus 5 is a really bad model\""}, {"id": 785288338, "item_id": 49080521, "url": "https://news.ycombinator.com/item?id=49080521", "by": "ignoramous", "date": "2026-07-28", "version": "4", "stance": "recommend", "quote": "I've found this to be true for 1M context-size open weights that I use (GLM 5.2, MiMo v2.5 Pro, & DeepSeek v4 Pro)", "excerpt": "> have terra write code and it will find lots of bugs in its own code. But let it iterate code reviews/fixes/test cases a few times, and you\u2019ll have something nice I've found this to be true for 1M context-size open weights that I use (GLM 5.2, MiMo v2.5 Pro, & DeepSeek v4 Pro). I think the prompt \"code review\", \"edge case\" etc triggers different parts of the LLM, bringing more files into its large context window. Th", "context": "", "use_case": "iterative code review and fixes", "switched_from": "", "story": "A $500 RL fine-tune of a 9B open model beat frontier models on catalog review"}, {"id": 785287584, "item_id": 49080474, "url": "https://news.ycombinator.com/item?id=49080474", "by": "solarkraft", "date": "2026-07-28", "version": "4", "stance": "recommend", "quote": "", "excerpt": "> even the lower-end models can find the things they didn't think of first when inspecting the already-written output I use Deepseek V4 Pro for my hobby project (an OpenCode client). The economics will obviously be different at work. It\u2019s worse than the GPTs, but indeed, if you focus it on reviewing its own code (essentially spending more reasoning and changing perspective), it\u2019s also quite capable at improving its o", "context": "hobby project developer", "use_case": "hobby OpenCode client development", "switched_from": "", "story": "A $500 RL fine-tune of a 9B open model beat frontier models on catalog review"}, {"id": 784795040, "item_id": 49049690, "url": "https://news.ycombinator.com/item?id=49049690", "by": "neongreen", "date": "2026-07-25", "version": "4", "stance": "neutral", "quote": "Deepseek v4 Pro has three modes: no thinking, thinking (\"high\"), very good thinking (\"max\").", "excerpt": "This is a very good question and the answers are different for different models. --- Deepseek v4 Pro has three modes: no thinking, thinking (\"high\"), very good thinking (\"max\"). Max is implemented as follows: https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/encoding/README.md#reasoning-effort > When reasoning_effort=\"max\" is set, a special prefix is prepended at the very beginning of the prompt (before th", "context": "", "use_case": "", "switched_from": "", "story": "What happens behind the scenes when we change effort for same LLM models?"}, {"id": 784220000, "item_id": 49013750, "url": "https://news.ycombinator.com/item?id=49013750", "by": "jmyeet", "date": "2026-07-22", "version": "4", "stance": "neutral", "quote": "The A100 came out in 2020 and cannot run a large model like DeepSeek v4 Pro.", "excerpt": "This is a good article. I've been curious about how this is going to play out. A couple of data points: 1. An enthusiast had a project to get a V100 working on his PC [1]. This was a ~$10k GPU 10 years ago. It's now sold for scrap; 2. The A100 came out in 2020 and cannot run a large model like DeepSeek v4 Pro. It can run Flash. You need a 16xH100 cluster to run Pro and that's a ~4 year old GPU and AFAICT 8xB100 or 4x", "context": "", "use_case": "", "switched_from": "", "story": "Nobody knows what a used GPU cluster is worth"}, {"id": 784014896, "item_id": 49000931, "url": "https://news.ycombinator.com/item?id=49000931", "by": "dools", "date": "2026-07-21", "version": "4", "stance": "recommend", "quote": "Yeah deepseek is my workhorse of choice these days", "excerpt": "Yeah deepseek is my workhorse of choice these days, I mostly use pro because I do a lot of concurrent jobs so speed is less of a concern. When it falls over I change the model mid-chat to GPT5.5 and chuck a couple of tokens over to OpenAI and then once it's correctly found the problem I switch back to deepseek, keeping all the 5.5 analysis in context. I have been a heavy user of k2.5/6 but they seem to have gotten sl", "context": "", "use_case": "concurrent coding jobs", "switched_from": "", "story": "Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA"}, {"id": 783856689, "item_id": 48991043, "url": "https://news.ycombinator.com/item?id=48991043", "by": "Squeeze2664", "date": "2026-07-21", "version": "4", "stance": "recommend", "quote": "It's also unbelievably cheap - so cheap, in fact, that I have started using it even for trivial tasks.", "excerpt": "I heavily use the Pi harness with DeepSeek-v4-pro directly from the DeepSeek API almost exclusively and I have have barely ever felt the need for something different. I have some OpenRouter credits and API key ready to go if DeepSeek fails me, and that almost never happens. It's also unbelievably cheap - so cheap, in fact, that I have started using it even for trivial tasks. I use it for coding, code review, managing", "context": "", "use_case": "coding, code review, general questions", "switched_from": "Claude Code", "story": "Ask HN: Claude Code or Codex?"}, {"id": 783050022, "item_id": 48940626, "url": "https://news.ycombinator.com/item?id=48940626", "by": "KronisLV", "date": "2026-07-16", "version": "4", "stance": "warn", "quote": "DeepSeek V4 Pro is really affordable per-token but regularly kept making mistakes in the tasks I gave it.", "excerpt": "> I've yet to find a way to run it on a monthly sub reliably for cheaper than Codex. Matches my experience, I got their Pro subscription and while I enjoyed the model itself a lot and while their ZCode harness is also pretty nice, it gave me less tokens for similar amounts of money that Anthropic would give me on a subscription: https://blog.kronis.dev/blog/z-ai-s-glm-5-2-is-a-great-model-but-is-it-good-value/ I'm ye", "context": "", "use_case": "", "switched_from": "", "story": "Kimi K3 is now live"}, {"id": 783005152, "item_id": 48937822, "url": "https://news.ycombinator.com/item?id=48937822", "by": "ignoramous", "date": "2026-07-16", "version": "4", "stance": "recommend", "quote": "are genuinely good enough as daily drivers, given the costs", "excerpt": "Possible, but pay-as-you-go Hy3 / DeepSeek v4 Pro / MiMo v2.5 Pro (from respective vendors) are genuinely good enough as daily drivers, given the costs (especially, low prices for input cache, which usually makes up 70%+ of total input for agentic workflows). I put in $10 in DeepSeek & Xiaomi MiMo, and I've barely used $1 each, in a week of coding work. Coding Plans by MiniMax ($20/mo for 1.7b tokens) and Z.ai (~$30/", "context": "", "use_case": "daily coding", "switched_from": "", "story": "Kimi K3 is now live"}, {"id": 781350561, "item_id": 48834410, "url": "https://news.ycombinator.com/item?id=48834410", "by": "wongarsu", "date": "2026-07-08", "version": "4", "stance": "recommend", "quote": "For hobby projects I've completely switched to DeepSeek v4 pro.", "excerpt": "At work I wouldn't want to use anything else. Compared to my salary a Claude subscription (or two) is cheap For hobby projects I've completely switched to DeepSeek v4 pro. I spend less than on a $10 Claude plan and am not subjected to quota limits (when I have time and motivation, the last thing I want is a 5 hour quota running out). And the difference in model performance is fine for those smaller projects, most of ", "context": "", "use_case": "hobby projects", "switched_from": "Claude", "story": "SWE-1.7 Reach Near GPT 5.5 and Opus Intelligence"}]}