{"slug": "mimo", "name": "MiMo", "category": "ai", "category_name": "AI models and assistants", "description": "Xiaomi's family of open-weight large language models, offered in Pro and Flash variants.", "recommenders": 23, "commenters": 37, "warners": 4, "score": 19.045, "rising": 13.71, "first": "2026-06-09", "latest": "2026-10-28", "why": "Liked as a cheap open model for coding work in agents and harnesses like OpenCode, with the Pro variant singled out for strong reasoning.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 23, "neutral": 11, "compare": 2, "warn": 3}, "spark": [0, 0, 0, 0, 0, 0, 0, 4, 2, 2, 8, 7], "listed": true, "loved": ["Low cost", "Coding capability"], "caveat": "Slow, overthinks", "short_quote": "We switched to Mimo 2.6, which is relatively better.", "icon": true, "domain": "xiaomi.com", "github_repo": null, "official_url": "https://mimo.xiaomi.com", "pricing": "open source", "aliases": ["MiMo", "MiMo 2.5", "MiMo v2.5 Pro", "MiMo 2.6", "MiMo V2.5 Pro", "MiMo v2.6", "Mimo", "MiMo-2.6", "MiMo 2.6 Pro", "MiMo-V2.5", "MiMo V2.5", "MiMo-2.5-pro", "MiMo-V2.6-Flash", "MiMo v2.6 Flash", "MiMo-V2.5-Pro", "MiMo-2.5-Pro", "MiMo v2.5", "MiMo 2.5 Pro", "Mimo 2.6 Pro", "Mimo V2.5 Pro"], "verdict": [{"text": "Liked as a cheap open model for coding work in agents and harnesses like OpenCode,", "cites": [{"id": 778794865, "item_id": 48674679, "url": "https://news.ycombinator.com/item?id=48674679", "by": "bel8", "date": "2026-06-25", "stance": "recommend", "quote": "You'll get 10x more value by using MiMo 2.5 or DeepSeek Flash which is what I use for most code.", "excerpt": "GLM 5.2 is priced at 1/3 of Opus. Depending on who you ask for a napkin math. Not worth it in my opinion. You'll get 10x more value by using MiMo 2.5 or DeepSeek Flash which is what I use for most code.", "context": "", "use_case": "coding", "switched_from": "", "story": "Elevated error rate across multiple models"}, {"id": 776119664, "item_id": 48507479, "url": "https://news.ycombinator.com/item?id=48507479", "by": "mdasen", "date": "2026-06-12", "stance": "recommend", "quote": "cached inputs are practically free", "excerpt": "I find that I don't use a ton of output tokens. I'm usually around 95% cached input, 4% input, and 1% output. For me, the big thing with MiMo-V2.5-Pro and DeepSeek V4-Pro is that cached inputs are practically free. Kimi K2.7 Code is 53x more expensive for cached inputs which is 95% of my costs. If I use 95M cached input tokens, 4M input tokens, and 1M output tokens, that'd be: $18 for cached input on Kimi K2.7 Code v", "context": "heavy cached-input API user", "use_case": "AI coding", "switched_from": "", "story": "Kimi K2.7-Code: open-source coding model with better token efficiency"}, {"id": 790838210, "item_id": 49427388, "url": "https://news.ycombinator.com/item?id=49427388", "by": "larsfaye", "date": "2026-08-24", "stance": "recommend", "quote": "Zed + OpenCode + Mimo/GLM/Kimi models are incredible. I haven't touched a \"frontier model\" for months.", "excerpt": "This is EXACTLY how I am working with these tools, you summarized everything about why I use this approach perfectly. Zed + OpenCode + Mimo/GLM/Kimi models are incredible. I haven't touched a \"frontier model\" for months. I really treat them as \"delegation layer\". They're systems that sit right alongside your stack, and at this point, at every level in it, like a connective tissue that intersects at all points. I can ", "context": "", "use_case": "AI coding", "switched_from": "", "story": "Coding expertise is going to collapse from AI reliance"}, {"id": 803831108, "item_id": 50239444, "url": "https://news.ycombinator.com/item?id=50239444", "by": "qlte", "date": "2026-10-26", "stance": "recommend", "quote": "I am entirely satisfied when using my Opencode Go plan with open models like Deepseek Flash/Pro, Kimi and MiMo.", "excerpt": "The profit comes from the enterprise customers who have to pay for usage at API rates. Not the subsidized subscriptions, which both OpenAI have designed to be unavailable for large enterprises by withholding IAM and auditing features they need except on the usage based entperise plans. As an individual I'm still paying for a ChatGPT Plus plan since $20/mo gives me a lot of value and I use many multiples of $20 at API", "context": "", "use_case": "general AI tasks", "switched_from": "", "story": "OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?"}]}, {"text": " with the Pro variant singled out for strong reasoning.", "cites": [{"id": 802501106, "item_id": 50156319, "url": "https://news.ycombinator.com/item?id=50156319", "by": "segmondy", "date": "2026-10-19", "stance": "recommend", "quote": "MiMo2.6-Pro is a beast", "excerpt": "Too early to call, I ran it through the weekend and compared with Qwen3.8-Flash-Next, DSv41F-1017 at FP8 and Q3.8FN at Q8. Both were great, but surprisingly, Q3.8FN had the better result. I was modelling and generating simulation code for a complex system. I used both because I had a bunch of images that needed to be used for input to steer the output. I think these models have strengths and it's best to experiment a", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4.1 Flash 1017 (Max)"}, {"id": 800236400, "item_id": 50014775, "url": "https://news.ycombinator.com/item?id=50014775", "by": "codeprimate", "date": "2026-10-09", "stance": "recommend", "quote": "Mimo 2.6 Pro is even better.", "excerpt": "Mimo 2.6 Pro is even better. I switched from DeepSeek 4.1 flash about 2 weeks ago for my Hermes sysadmin/coding agents and I am seeing better intelligence and lower overall spend. https://artificialanalysis.ai/models/mimo-v2-6-pro", "context": "", "use_case": "sysadmin and coding agents", "switched_from": "DeepSeek 4.1 Flash", "story": "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"}, {"id": 796708912, "item_id": 49794307, "url": "https://news.ycombinator.com/item?id=49794307", "by": "GodelNumbering", "date": "2026-09-21", "stance": "recommend", "quote": "Mimo has been one of those models that I have been rooting for since the first I used it", "excerpt": "Mimo has been one of those models that I have been rooting for since the first I used it, the 2.5 pro which I have used quite a bit, was very concise, very aware of how much context needs to be read for which tasks and would always keep the context tight. Also surprisingly good at strategic thinking. I had published a comparison between it and Terra where Terra was found to be using much more avg context for similar ", "context": "", "use_case": "agentic tasks", "switched_from": "", "story": "MiMo v2.6"}]}, {"text": " The main complaints are slow responses and overthinking,", "cites": [{"id": 802516449, "item_id": 50157278, "url": "https://news.ycombinator.com/item?id=50157278", "by": "sinuhe69", "date": "2026-10-19", "stance": "warn", "quote": "all Mino models are painfully slow", "excerpt": "On Openrouter at least, all Mino models are painfully slow. TTFT is many seconds and the models spend a lot of time on its CoT, making effective output per second many times slower than DS Flash. On Flash 4.1, I constantly get ~180 tok/s and the real work got done fast, too. So even when the Mimo 2.6 are smart models, I use them much less recently. Not to mention their costs are also higher than DS. For me, the DeepS", "context": "", "use_case": "", "switched_from": "", "story": "MiniMax M3.1 Flash"}, {"id": 804330772, "item_id": 50270673, "url": "https://news.ycombinator.com/item?id=50270673", "by": "wyrdcurt", "date": "2026-10-28", "stance": "warn", "quote": "I've personally experienced overthinking and the thinking loop issue more often than seeing it not think enough", "excerpt": "I've personally experienced overthinking and the thinking loop issue more often than seeing it not think enough, I know what you mean though. I tried Haiku 5.5 and the most impressive thing about it to me was that it seemed not to overthink when it wasn't necessary, yet it somehow knew when it should spend some more time reasoning about a task. American models are too rich for my blood though. I use GLM 5.3 Flash mor", "context": "", "use_case": "", "switched_from": "", "story": "Artificial Analysis Intelligence Index v5"}]}, {"text": " and some find it uneven at following instructions.", "cites": [{"id": 785791600, "item_id": 49111975, "url": "https://news.ycombinator.com/item?id=49111975", "by": "CapyToolkit", "date": "2026-07-30", "stance": "warn", "quote": "All behave dumb, don't follow instructions properly even when they write skills for themselves.", "excerpt": "Honestly, I have a similar impressions after testing every model except Sonnet, Opus or Claude. I always test them inside Claude vscode extension in the terminal, using the same skills or workflows from multiple skills. I tested Mimo V2.5 Pro, Qwen 3.8 Max Preview, GLM 5.2, Macaron V1 Venti and many other, smaller model whenever hype on X appears. All behave dumb, don't follow instructions properly even when they wri", "context": "tests many models", "use_case": "simple tasks", "switched_from": "", "story": "Kimi-code not performing well"}, {"id": 802679061, "item_id": 50167441, "url": "https://news.ycombinator.com/item?id=50167441", "by": "A_D_E_P_T", "date": "2026-10-20", "stance": "recommend", "quote": "You should also test MiMo v2.6.", "excerpt": "The Google models are great \"general purpose\" and research LLMs, but they're relatively poor at coding. The way I see it, you have two options. 1) Get on the Claude Max 20x plan and use Opus 5.5 for review/orchestration and Haiku 5.5 for grunt work. Haiku 5.5 is decent, cheap, and fast. 2) Deepseek 4.1 Flash in OpenCode. You should also test MiMo v2.6. It's a little bit uneven, but might work well for whatever you're", "context": "", "use_case": "cheap AI coding", "switched_from": "", "story": "Ask HN: Best value/money AI coding plan bundle?"}]}], "praise": [{"label": "Low cost", "summary": "Commenters say it gets a lot of work done for little money, with near-free cached input tokens.", "quotes": [{"id": 776119664, "item_id": 48507479, "url": "https://news.ycombinator.com/item?id=48507479", "by": "mdasen", "date": "2026-06-12", "stance": "recommend", "quote": "cached inputs are practically free", "excerpt": "I find that I don't use a ton of output tokens. I'm usually around 95% cached input, 4% input, and 1% output. For me, the big thing with MiMo-V2.5-Pro and DeepSeek V4-Pro is that cached inputs are practically free. Kimi K2.7 Code is 53x more expensive for cached inputs which is 95% of my costs. If I use 95M cached input tokens, 4M input tokens, and 1M output tokens, that'd be: $18 for cached input on Kimi K2.7 Code v", "context": "heavy cached-input API user", "use_case": "AI coding", "switched_from": "", "story": "Kimi K2.7-Code: open-source coding model with better token efficiency"}, {"id": 778794865, "item_id": 48674679, "url": "https://news.ycombinator.com/item?id=48674679", "by": "bel8", "date": "2026-06-25", "stance": "recommend", "quote": "You'll get 10x more value by using MiMo 2.5 or DeepSeek Flash which is what I use for most code.", "excerpt": "GLM 5.2 is priced at 1/3 of Opus. Depending on who you ask for a napkin math. Not worth it in my opinion. You'll get 10x more value by using MiMo 2.5 or DeepSeek Flash which is what I use for most code.", "context": "", "use_case": "coding", "switched_from": "", "story": "Elevated error rate across multiple models"}, {"id": 796942497, "item_id": 49808906, "url": "https://news.ycombinator.com/item?id=49808906", "by": "system2", "date": "2026-09-22", "stance": "recommend", "quote": "We switched to Mimo 2.6, which is relatively better.", "excerpt": "Except for the censorship. We use it for massive data crunching, and roughly 5-8% (depending on the day) gets censored and doesn't get a response. We switched to Mimo 2.6, which is relatively better. For censored stuff, we use Sonnet and OpenAI Nano models. Also Mimo 2.6 is roughly 30% cheaper. Without batch.", "context": "", "use_case": "massive data crunching", "switched_from": "GLM 5.3", "story": "GPT-6 Sol and Luna"}]}, {"label": "Coding capability", "summary": "Several use it as a daily coding model in place of frontier models, including on large codebases.", "quotes": [{"id": 790838210, "item_id": 49427388, "url": "https://news.ycombinator.com/item?id=49427388", "by": "larsfaye", "date": "2026-08-24", "stance": "recommend", "quote": "Zed + OpenCode + Mimo/GLM/Kimi models are incredible. I haven't touched a \"frontier model\" for months.", "excerpt": "This is EXACTLY how I am working with these tools, you summarized everything about why I use this approach perfectly. Zed + OpenCode + Mimo/GLM/Kimi models are incredible. I haven't touched a \"frontier model\" for months. I really treat them as \"delegation layer\". They're systems that sit right alongside your stack, and at this point, at every level in it, like a connective tissue that intersects at all points. I can ", "context": "", "use_case": "AI coding", "switched_from": "", "story": "Coding expertise is going to collapse from AI reliance"}, {"id": 779515825, "item_id": 48719739, "url": "https://news.ycombinator.com/item?id=48719739", "by": "bel8", "date": "2026-06-29", "stance": "recommend", "quote": "I use DeepSeek V4 Flash (high) and MiMo 2.5 (non Pro, because vision)", "excerpt": "Why would you say that? I use DeepSeek V4 Flash (high) and MiMo 2.5 (non Pro, because vision) to work on medium sized projects (~1mil lines of code, C#, Go, TypeScript) with great success. And that is coming from someone who used Opus 4.7 and GPT 5.5 as workhorses before. And I'm pretty sure GLM 5.2 is better than the lighter models I use. My worflow is simple: plan -> clarify -> implement. 1) plan prompt template: I", "context": "works on ~1M line codebases", "use_case": "coding with vision", "switched_from": "Opus 4.7", "story": "Semgrep: GLM 5.2 beats Claude in our Cyber Benchmarks"}, {"id": 796988000, "item_id": 49811750, "url": "https://news.ycombinator.com/item?id=49811750", "by": "sieve", "date": "2026-09-23", "stance": "recommend", "quote": "perfectly capable of following my spec and then poking holes in the implementation", "excerpt": "I used to read the code till around May. Now I don't. Instead I validate behavior. And have multiple LLMs verify that the code implements my handwritten spec. MiMo 2.5/2.6, MuseSpark 1.3, DeepSeek V4/4.1 Flash and GLM 5.3 Flash are perfectly capable of following my spec and then poking holes in the implementation till there are none left.", "context": "validates behavior instead of reading code", "use_case": "spec implementation and verification", "switched_from": "", "story": "GPT-6 Sol and Luna"}]}, {"label": "Pro reasoning", "summary": "The Pro variant is described as very capable, with better intelligence than DeepSeek Flash for agent work.", "quotes": [{"id": 802501106, "item_id": 50156319, "url": "https://news.ycombinator.com/item?id=50156319", "by": "segmondy", "date": "2026-10-19", "stance": "recommend", "quote": "MiMo2.6-Pro is a beast", "excerpt": "Too early to call, I ran it through the weekend and compared with Qwen3.8-Flash-Next, DSv41F-1017 at FP8 and Q3.8FN at Q8. Both were great, but surprisingly, Q3.8FN had the better result. I was modelling and generating simulation code for a complex system. I used both because I had a bunch of images that needed to be used for input to steer the output. I think these models have strengths and it's best to experiment a", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4.1 Flash 1017 (Max)"}, {"id": 800236400, "item_id": 50014775, "url": "https://news.ycombinator.com/item?id=50014775", "by": "codeprimate", "date": "2026-10-09", "stance": "recommend", "quote": "Mimo 2.6 Pro is even better.", "excerpt": "Mimo 2.6 Pro is even better. I switched from DeepSeek 4.1 flash about 2 weeks ago for my Hermes sysadmin/coding agents and I am seeing better intelligence and lower overall spend. https://artificialanalysis.ai/models/mimo-v2-6-pro", "context": "", "use_case": "sysadmin and coding agents", "switched_from": "DeepSeek 4.1 Flash", "story": "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"}, {"id": 775328145, "item_id": 48458009, "url": "https://news.ycombinator.com/item?id=48458009", "by": "ricardobeat", "date": "2026-06-09", "stance": "neutral", "quote": "I use Pro only for one-off tasks that require complex reasoning", "excerpt": "Mimo v2.5 non-pro seems to do better with tool usage than its Pro sibling, is much cheaper and solves 90% of the same problems. I use Pro only for one-off tasks that require complex reasoning: memory management bugs, algorithms, planning. When it gets stuck, I get one-shot advice from Claude or DS Pro. I\u2019ve done massive amounts of work for cheap this way.", "context": "", "use_case": "complex reasoning tasks", "switched_from": "", "story": "MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second"}]}, {"label": "Concise output", "summary": "Users note it keeps context tight and is less chatty, with a refreshing communication style.", "quotes": [{"id": 796708912, "item_id": 49794307, "url": "https://news.ycombinator.com/item?id=49794307", "by": "GodelNumbering", "date": "2026-09-21", "stance": "recommend", "quote": "Mimo has been one of those models that I have been rooting for since the first I used it", "excerpt": "Mimo has been one of those models that I have been rooting for since the first I used it, the 2.5 pro which I have used quite a bit, was very concise, very aware of how much context needs to be read for which tasks and would always keep the context tight. Also surprisingly good at strategic thinking. I had published a comparison between it and Terra where Terra was found to be using much more avg context for similar ", "context": "", "use_case": "agentic tasks", "switched_from": "", "story": "MiMo v2.6"}, {"id": 783407777, "item_id": 48962986, "url": "https://news.ycombinator.com/item?id=48962986", "by": "ignoramous", "date": "2026-07-18", "stance": "recommend", "quote": "Prefer the latter over DeepSeek v4 Pro because it costs the same while being equally good & less chattier for coding workloads.", "excerpt": "> It's cheap enough to where I don't think about cost, made even cheaper by DeepSeek having the most effective and cheap caching in the industry I use DeepSeek v4 Flash & MiMo v2.5 Pro. Prefer the latter over DeepSeek v4 Pro because it costs the same while being equally good & less chattier for coding workloads. Although, I've begun experimenting with Hy3 (as an in-between Flash & Pro) & GLM 5.2 (for long-horizon tas", "context": "", "use_case": "coding workloads", "switched_from": "", "story": "The Kimi K3 Moment"}, {"id": 804323312, "item_id": 50270207, "url": "https://news.ycombinator.com/item?id=50270207", "by": "wyre", "date": "2026-10-28", "stance": "recommend", "quote": "I really liked it, its communication style was a breath of fresh air compared to how stale gpt-sol is.", "excerpt": "I used MiMo for a couple of days waiting for my Codex sub to reset. I really liked it, its communication style was a breath of fresh air compared to how stale gpt-sol is.", "context": "", "use_case": "coding while Codex limit reset", "switched_from": "", "story": "Artificial Analysis Intelligence Index v5"}]}, {"label": "Training transparency", "summary": "Commenters praise Xiaomi's openness about training methodology, benchmarks and release notes.", "quotes": [{"id": 796687968, "item_id": 49792998, "url": "https://news.ycombinator.com/item?id=49792998", "by": "rao-v", "date": "2026-09-21", "stance": "recommend", "quote": "I really like how transparent they\u2019ve been about the training of this model.", "excerpt": "I know we have strong views on what a truly open model is (open weights, open training data, open training code etc.) but I really like how transparent they\u2019ve been about the training of this model. The realtime dashboard they shared during training (https://mimo.xiaomi.com/rl/) was an incredible learning and teaching tool for me, and they\u2019ve been unusually comprehensive in sharing details about their methodology (ch", "context": "", "use_case": "open model", "switched_from": "", "story": "MiMo v2.6"}, {"id": 796884384, "item_id": 49805274, "url": "https://news.ycombinator.com/item?id=49805274", "by": "peri-cl", "date": "2026-09-22", "stance": "recommend", "quote": "I love the contrast with yesterday's open-source MiMo release", "excerpt": "I love the contrast with yesterday's open-source MiMo release, which put research chemistry (metal-organic frameworks stuff) front and center in the release notes. https://mimo.xiaomi.com/mimo-v2-6#co-scientist-for-materials-research", "context": "", "use_case": "research chemistry", "switched_from": "", "story": "Claude Opus 5.5"}]}, {"label": "Fewer refusals", "summary": "One team switched to it from GLM because it censored fewer bulk data-processing requests.", "quotes": [{"id": 796942497, "item_id": 49808906, "url": "https://news.ycombinator.com/item?id=49808906", "by": "system2", "date": "2026-09-22", "stance": "recommend", "quote": "We switched to Mimo 2.6, which is relatively better.", "excerpt": "Except for the censorship. We use it for massive data crunching, and roughly 5-8% (depending on the day) gets censored and doesn't get a response. We switched to Mimo 2.6, which is relatively better. For censored stuff, we use Sonnet and OpenAI Nano models. Also Mimo 2.6 is roughly 30% cheaper. Without batch.", "context": "", "use_case": "massive data crunching", "switched_from": "GLM 5.3", "story": "GPT-6 Sol and Luna"}]}], "caveats": [{"label": "Slow, overthinks", "summary": "Users report long time-to-first-token, lengthy chain-of-thought and thinking loops.", "quotes": [{"id": 802516449, "item_id": 50157278, "url": "https://news.ycombinator.com/item?id=50157278", "by": "sinuhe69", "date": "2026-10-19", "stance": "warn", "quote": "all Mino models are painfully slow", "excerpt": "On Openrouter at least, all Mino models are painfully slow. TTFT is many seconds and the models spend a lot of time on its CoT, making effective output per second many times slower than DS Flash. On Flash 4.1, I constantly get ~180 tok/s and the real work got done fast, too. So even when the Mimo 2.6 are smart models, I use them much less recently. Not to mention their costs are also higher than DS. For me, the DeepS", "context": "", "use_case": "", "switched_from": "", "story": "MiniMax M3.1 Flash"}, {"id": 804330772, "item_id": 50270673, "url": "https://news.ycombinator.com/item?id=50270673", "by": "wyrdcurt", "date": "2026-10-28", "stance": "warn", "quote": "I've personally experienced overthinking and the thinking loop issue more often than seeing it not think enough", "excerpt": "I've personally experienced overthinking and the thinking loop issue more often than seeing it not think enough, I know what you mean though. I tried Haiku 5.5 and the most impressive thing about it to me was that it seemed not to overthink when it wasn't necessary, yet it somehow knew when it should spend some more time reasoning about a task. American models are too rich for my blood though. I use GLM 5.3 Flash mor", "context": "", "use_case": "", "switched_from": "", "story": "Artificial Analysis Intelligence Index v5"}]}, {"label": "Uneven quality", "summary": "Some find it uneven or poor at following instructions, and the Flash variant weaker than Pro.", "quotes": [{"id": 785791600, "item_id": 49111975, "url": "https://news.ycombinator.com/item?id=49111975", "by": "CapyToolkit", "date": "2026-07-30", "stance": "warn", "quote": "All behave dumb, don't follow instructions properly even when they write skills for themselves.", "excerpt": "Honestly, I have a similar impressions after testing every model except Sonnet, Opus or Claude. I always test them inside Claude vscode extension in the terminal, using the same skills or workflows from multiple skills. I tested Mimo V2.5 Pro, Qwen 3.8 Max Preview, GLM 5.2, Macaron V1 Venti and many other, smaller model whenever hype on X appears. All behave dumb, don't follow instructions properly even when they wri", "context": "tests many models", "use_case": "simple tasks", "switched_from": "", "story": "Kimi-code not performing well"}, {"id": 802679061, "item_id": 50167441, "url": "https://news.ycombinator.com/item?id=50167441", "by": "A_D_E_P_T", "date": "2026-10-20", "stance": "recommend", "quote": "You should also test MiMo v2.6.", "excerpt": "The Google models are great \"general purpose\" and research LLMs, but they're relatively poor at coding. The way I see it, you have two options. 1) Get on the Claude Max 20x plan and use Opus 5.5 for review/orchestration and Haiku 5.5 for grunt work. Haiku 5.5 is decent, cheap, and fast. 2) Deepseek 4.1 Flash in OpenCode. You should also test MiMo v2.6. It's a little bit uneven, but might work well for whatever you're", "context": "", "use_case": "cheap AI coding", "switched_from": "", "story": "Ask HN: Best value/money AI coding plan bundle?"}, {"id": 802501106, "item_id": 50156319, "url": "https://news.ycombinator.com/item?id=50156319", "by": "segmondy", "date": "2026-10-19", "stance": "recommend", "quote": "MiMo2.6-Pro is a beast", "excerpt": "Too early to call, I ran it through the weekend and compared with Qwen3.8-Flash-Next, DSv41F-1017 at FP8 and Q3.8FN at Q8. Both were great, but surprisingly, Q3.8FN had the better result. I was modelling and generating simulation code for a complex system. I used both because I had a bunch of images that needed to be used for input to steer the output. I think these models have strengths and it's best to experiment a", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4.1 Flash 1017 (Max)"}]}, {"label": "Accepts regressions", "summary": "In one comparison it pushed the user to accept a functional regression instead of fixing it.", "quotes": [{"id": 778697187, "item_id": 48668574, "url": "https://news.ycombinator.com/item?id=48668574", "by": "JSR_FDED", "date": "2026-06-24", "stance": "compare", "quote": "Deepseek, Kimi, MiMo started giving me multiple options but advocating strongly that I should either accept this deficiency", "excerpt": "Anecdote, not \u201cresearch\u201d: Yesterday I compared Deepseek, Kimi 2.6, MiMo 2.5 and GLM 5.2 for the same task (replace a custom token-based auth scheme with a cookies-based scheme across a front- and back-end codebase). I used Opencode with the zen subscription to try different models. All did this perfectly, basically indistinguishable from each other. However, when I pointed out that the new cookies-based auth didn\u2019t a", "context": "", "use_case": "coding", "switched_from": "", "story": "GLM-5.2 is the step change for open agents"}]}, {"label": "Data guarantees", "summary": "One commenter questions the data-training guarantees of the available hosting options for company use.", "quotes": [{"id": 796712656, "item_id": 49794541, "url": "https://news.ycombinator.com/item?id=49794541", "by": "wkcheng", "date": "2026-09-21", "stance": "neutral", "quote": "Are people using MiMo models as their daily drivers in a company setting?", "excerpt": "Are people using MiMo models as their daily drivers in a company setting? If so, how? I know they're available over OpenCode and directly from Xiaomi, but those are not great options. OpenCode Go straight up doesn't give any guarantees about training on your data, and Xiaomi says that they won't but it's unclear. With some models you can find hosting companies based in the EU or US, but then you don't know how they'r", "context": "", "use_case": "company daily driver model", "switched_from": "", "story": "MiMo v2.6"}]}, {"label": "Confusing plans", "summary": "Its token plans are described as a good deal but confusing.", "quotes": [{"id": 780015572, "item_id": 48750973, "url": "https://news.ycombinator.com/item?id=48750973", "by": "SwellJoe", "date": "2026-07-01", "stance": "neutral", "quote": "MiMo has token plans that are a pretty good deal (though confusing", "excerpt": "North Mini Code is on par with the similar-sized MoE Qwen 3.6 and Gemma 4. It benches a little better than Gemma 4 26b a4b and a little worse than Qwen 3.6 35b a3b for most coding-related and agentic tasks. For my use case, or at least the use case I've tested, it is a little worse than both for surfacing security bugs, but all of the MoE models at this size are pretty bad at finding security bugs (they hallucinate a", "context": "", "use_case": "token plans", "switched_from": "", "story": "Qwen 3.6 27B is the sweet spot for local development"}]}, {"label": "Pro tool use", "summary": "One user found the non-Pro model handled tool usage better than Pro.", "quotes": [{"id": 775328144, "item_id": 48458009, "url": "https://news.ycombinator.com/item?id=48458009", "by": "ricardobeat", "date": "2026-06-09", "stance": "recommend", "quote": "Mimo v2.5 non-pro seems to do better with tool usage than its Pro sibling, is much cheaper and solves 90% of the same problems.", "excerpt": "Mimo v2.5 non-pro seems to do better with tool usage than its Pro sibling, is much cheaper and solves 90% of the same problems. I use Pro only for one-off tasks that require complex reasoning: memory management bugs, algorithms, planning. When it gets stuck, I get one-shot advice from Claude or DS Pro. I\u2019ve done massive amounts of work for cheap this way.", "context": "", "use_case": "cheap coding tasks with tool use", "switched_from": "", "story": "MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second"}]}], "who": [{"by": "abalashov", "context": "individual contributor", "url": "https://news.ycombinator.com/item?id=49815221", "date": "2026-09-23"}, {"by": "sieve", "context": "validates behavior instead of reading code", "url": "https://news.ycombinator.com/item?id=49811750", "date": "2026-09-23"}, {"by": "vardalab", "context": "homelab user", "url": "https://news.ycombinator.com/item?id=49805228", "date": "2026-09-22"}, {"by": "surgical_fire", "context": "uses Sonnet and Opus at work", "url": "https://news.ycombinator.com/item?id=49174505", "date": "2026-08-04"}, {"by": "bel8", "context": "works on ~1M line codebases", "url": "https://news.ycombinator.com/item?id=48719739", "date": "2026-06-29"}, {"by": "trollbridge", "context": "small dev team of 6", "url": "https://news.ycombinator.com/item?id=48648590", "date": "2026-06-23"}, {"by": "maxrev17", "context": "built own agent harness", "url": "https://news.ycombinator.com/item?id=48589811", "date": "2026-06-18"}, {"by": "mdasen", "context": "heavy cached-input API user", "url": "https://news.ycombinator.com/item?id=48507479", "date": "2026-06-12"}], "tags": ["for coding agents", "with OpenCode", "for budget API users", "self-hostable", "for side projects"], "use_cases": ["coding", "ai coding", "coding with vision", "coding while codex limit reset", "general ai tasks", "coding with janet"], "compared": [{"slug": "deepseek", "name": "DeepSeek", "co": 26, "switched_from": 0, "switched_to": 0}, {"slug": "glm", "name": "GLM", "co": 21, "switched_from": 1, "switched_to": 0}, {"slug": "deepseek-v4-flash", "name": "DeepSeek V4 Flash", "co": 17, "switched_from": 1, "switched_to": 0}, {"slug": "opencode", "name": "OpenCode", "co": 15, "switched_from": 0, "switched_to": 0}, {"slug": "kimi", "name": "Kimi", "co": 10, "switched_from": 0, "switched_to": 0}, {"slug": "claude-opus", "name": "Claude Opus", "co": 7, "switched_from": 1, "switched_to": 0}, {"slug": "claude", "name": "Claude", "co": 8, "switched_from": 0, "switched_to": 0}, {"slug": "openrouter", "name": "OpenRouter", "co": 7, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=49792730", "title": "MiMo v2.6", "mentions": 5}, {"url": "https://news.ycombinator.com/item?id=49805509", "title": "GPT-6 Sol and Luna", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=50268868", "title": "Artificial Analysis Intelligence Index v5", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49803892", "title": "Claude Opus 5.5", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=48721903", "title": "Qwen 3.6 27B is the sweet spot for local development", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=48446639", "title": "MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second", "mentions": 2}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {}, "2026-04": {}, "2026-05": {}, "2026-06": {"recommend": 4, "neutral": 3, "compare": 1}, "2026-07": {"neutral": 3, "recommend": 2, "warn": 1}, "2026-08": {"recommend": 2, "compare": 1}, "2026-09": {"warn": 1, "recommend": 8, "neutral": 4}, "2026-10": {"recommend": 7, "neutral": 1, "warn": 1}}, "quotes": [{"id": 804330772, "item_id": 50270673, "url": "https://news.ycombinator.com/item?id=50270673", "by": "wyrdcurt", "date": "2026-10-28", "stance": "warn", "quote": "I've personally experienced overthinking and the thinking loop issue more often than seeing it not think enough", "excerpt": "I've personally experienced overthinking and the thinking loop issue more often than seeing it not think enough, I know what you mean though. I tried Haiku 5.5 and the most impressive thing about it to me was that it seemed not to overthink when it wasn't necessary, yet it somehow knew when it should spend some more time reasoning about a task. American models are too rich for my blood though. I use GLM 5.3 Flash mor", "context": "", "use_case": "", "switched_from": "", "story": "Artificial Analysis Intelligence Index v5"}, {"id": 804323312, "item_id": 50270207, "url": "https://news.ycombinator.com/item?id=50270207", "by": "wyre", "date": "2026-10-28", "stance": "recommend", "quote": "I really liked it, its communication style was a breath of fresh air compared to how stale gpt-sol is.", "excerpt": "I used MiMo for a couple of days waiting for my Codex sub to reset. I really liked it, its communication style was a breath of fresh air compared to how stale gpt-sol is.", "context": "", "use_case": "coding while Codex limit reset", "switched_from": "", "story": "Artificial Analysis Intelligence Index v5"}, {"id": 803831108, "item_id": 50239444, "url": "https://news.ycombinator.com/item?id=50239444", "by": "qlte", "date": "2026-10-26", "stance": "recommend", "quote": "I am entirely satisfied when using my Opencode Go plan with open models like Deepseek Flash/Pro, Kimi and MiMo.", "excerpt": "The profit comes from the enterprise customers who have to pay for usage at API rates. Not the subsidized subscriptions, which both OpenAI have designed to be unavailable for large enterprises by withholding IAM and auditing features they need except on the usage based entperise plans. As an individual I'm still paying for a ChatGPT Plus plan since $20/mo gives me a lot of value and I use many multiples of $20 at API", "context": "", "use_case": "general AI tasks", "switched_from": "", "story": "OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?"}, {"id": 803314627, "item_id": 50207164, "url": "https://news.ycombinator.com/item?id=50207164", "by": "phillc73", "date": "2026-10-23", "stance": "neutral", "quote": "a little bit of MiMo-2.6", "excerpt": "I use Maki[1] harness and GML-5.3, GLM5.3 Flash and Deepseek-4.1 Flash 1017, with a little bit of MiMo-2.6. The key I've found is a strong skill. At the beginning, sure there were problems. However, I implemented an AGENTS rule that said if a Janet script failed more than twice on the same \"gotcha\", the correct syntax was added to the skill. The skill is regularly reviewed, consolidated and pruned. The Janet skill is", "context": "", "use_case": "coding with Janet", "switched_from": "", "story": "Jank \u2013 The native Clojure dialect with seamless C++ interop"}, {"id": 802679061, "item_id": 50167441, "url": "https://news.ycombinator.com/item?id=50167441", "by": "A_D_E_P_T", "date": "2026-10-20", "stance": "recommend", "quote": "You should also test MiMo v2.6.", "excerpt": "The Google models are great \"general purpose\" and research LLMs, but they're relatively poor at coding. The way I see it, you have two options. 1) Get on the Claude Max 20x plan and use Opus 5.5 for review/orchestration and Haiku 5.5 for grunt work. Haiku 5.5 is decent, cheap, and fast. 2) Deepseek 4.1 Flash in OpenCode. You should also test MiMo v2.6. It's a little bit uneven, but might work well for whatever you're", "context": "", "use_case": "cheap AI coding", "switched_from": "", "story": "Ask HN: Best value/money AI coding plan bundle?"}, {"id": 802516449, "item_id": 50157278, "url": "https://news.ycombinator.com/item?id=50157278", "by": "sinuhe69", "date": "2026-10-19", "stance": "warn", "quote": "all Mino models are painfully slow", "excerpt": "On Openrouter at least, all Mino models are painfully slow. TTFT is many seconds and the models spend a lot of time on its CoT, making effective output per second many times slower than DS Flash. On Flash 4.1, I constantly get ~180 tok/s and the real work got done fast, too. So even when the Mimo 2.6 are smart models, I use them much less recently. Not to mention their costs are also higher than DS. For me, the DeepS", "context": "", "use_case": "", "switched_from": "", "story": "MiniMax M3.1 Flash"}, {"id": 802501106, "item_id": 50156319, "url": "https://news.ycombinator.com/item?id=50156319", "by": "segmondy", "date": "2026-10-19", "stance": "recommend", "quote": "MiMo2.6-Pro is a beast", "excerpt": "Too early to call, I ran it through the weekend and compared with Qwen3.8-Flash-Next, DSv41F-1017 at FP8 and Q3.8FN at Q8. Both were great, but surprisingly, Q3.8FN had the better result. I was modelling and generating simulation code for a complex system. I used both because I had a bunch of images that needed to be used for input to steer the output. I think these models have strengths and it's best to experiment a", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4.1 Flash 1017 (Max)"}, {"id": 800236400, "item_id": 50014775, "url": "https://news.ycombinator.com/item?id=50014775", "by": "codeprimate", "date": "2026-10-09", "stance": "recommend", "quote": "Mimo 2.6 Pro is even better.", "excerpt": "Mimo 2.6 Pro is even better. I switched from DeepSeek 4.1 flash about 2 weeks ago for my Hermes sysadmin/coding agents and I am seeing better intelligence and lower overall spend. https://artificialanalysis.ai/models/mimo-v2-6-pro", "context": "", "use_case": "sysadmin and coding agents", "switched_from": "DeepSeek 4.1 Flash", "story": "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"}, {"id": 799953681, "item_id": 49997105, "url": "https://news.ycombinator.com/item?id=49997105", "by": "wyrdcurt", "date": "2026-10-07", "stance": "neutral", "quote": "along with some MiMo and DeepSeek", "excerpt": "Some people/organizations are ideologically opposed to using Chinese models. Not me, I use GLM-5.3-Flash for almost everything (the subscription-subsidized pricing on a legacy Z.ai plan makes it the best value model by a wide margin), along with some MiMo and DeepSeek. Still, I use Luna for certain tasks where speed is more valuable than performance; I can see this new Haiku displacing Luna for those. If you mean Hai", "context": "", "use_case": "", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 799694546, "item_id": 49980909, "url": "https://news.ycombinator.com/item?id=49980909", "by": "phillc73", "date": "2026-10-06", "stance": "recommend", "quote": "So far so good, seem to have done a lot of work for the $2.85 I've spent", "excerpt": "I'm at a loss as to what to do now. I've been wanting to support Mistral for so long. I struggled on with Mistral Medium 3.5 for longer than I should have (although I also think it taught me some valuable process lessons). Recently I switched to the Mistral hosted GLM-5.3, this worked very well and powered through a tonne of work. Unfortunately, I also completely maxed out two subscriptions within the space of six da", "context": "", "use_case": "coding", "switched_from": "", "story": "Mistral Large 4"}, {"id": 799135010, "item_id": 49945938, "url": "https://news.ycombinator.com/item?id=49945938", "by": "surgical_fire", "date": "2026-10-03", "stance": "recommend", "quote": "I use Chinese models on those. GLM, DeepSeek, MiMo.", "excerpt": "On my small personal projects I seldom find any mistakes. I use Chinese models on those. GLM, DeepSeek, MiMo. At work I use Claude, but it is a considerably larger and quite complex codebase. I use LLMs a lot, but it is a common weekly occurrence for me to correct misconceptions, bugs, or overengineering from Opus/Sonnet (Opus plans and reviews, Sonnet implements).", "context": "", "use_case": "small personal projects coding", "switched_from": "", "story": "Zig v0.17.0"}, {"id": 797043540, "item_id": 49815221, "url": "https://news.ycombinator.com/item?id=49815221", "by": "abalashov", "date": "2026-09-23", "stance": "neutral", "quote": "GLM 5.2/5.3, and MiMo thrown in", "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": "coding", "switched_from": "", "story": "The current balance of power in open models"}, {"id": 796988000, "item_id": 49811750, "url": "https://news.ycombinator.com/item?id=49811750", "by": "sieve", "date": "2026-09-23", "stance": "recommend", "quote": "perfectly capable of following my spec and then poking holes in the implementation", "excerpt": "I used to read the code till around May. Now I don't. Instead I validate behavior. And have multiple LLMs verify that the code implements my handwritten spec. MiMo 2.5/2.6, MuseSpark 1.3, DeepSeek V4/4.1 Flash and GLM 5.3 Flash are perfectly capable of following my spec and then poking holes in the implementation till there are none left.", "context": "validates behavior instead of reading code", "use_case": "spec implementation and verification", "switched_from": "", "story": "GPT-6 Sol and Luna"}, {"id": 796972147, "item_id": 49810759, "url": "https://news.ycombinator.com/item?id=49810759", "by": "sieve", "date": "2026-09-23", "stance": "recommend", "quote": "DS4/MS13/MiMo are almost there for these use cases as well.", "excerpt": "I use Claude Sonnet and ChatGPT via the web UI. I often use Claude to come up with specs for my ideas. This is becoming less and less useful. DS4/MS13/MiMo are almost there for these use cases as well. I dogfood everything I produce, and the models are good at collaborating with me on a spec and then turning it into code. If Sonnet/ChatGPT suddenly became unavailable due to Anthropic/OpenAI suddenly not being able to", "context": "", "use_case": "specs and code", "switched_from": "", "story": "GPT-6 Sol and Luna"}, {"id": 796953684, "item_id": 49809605, "url": "https://news.ycombinator.com/item?id=49809605", "by": "abalashov", "date": "2026-09-22", "stance": "neutral", "quote": "GLM 5.2/5.3, and MiMo thrown in", "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": "", "switched_from": "", "story": "MiMo-v2.6-Pro: Intelligence, Performance and Price Analysis"}, {"id": 796942497, "item_id": 49808906, "url": "https://news.ycombinator.com/item?id=49808906", "by": "system2", "date": "2026-09-22", "stance": "recommend", "quote": "We switched to Mimo 2.6, which is relatively better.", "excerpt": "Except for the censorship. We use it for massive data crunching, and roughly 5-8% (depending on the day) gets censored and doesn't get a response. We switched to Mimo 2.6, which is relatively better. For censored stuff, we use Sonnet and OpenAI Nano models. Also Mimo 2.6 is roughly 30% cheaper. Without batch.", "context": "", "use_case": "massive data crunching", "switched_from": "GLM 5.3", "story": "GPT-6 Sol and Luna"}, {"id": 796884384, "item_id": 49805274, "url": "https://news.ycombinator.com/item?id=49805274", "by": "peri-cl", "date": "2026-09-22", "stance": "recommend", "quote": "I love the contrast with yesterday's open-source MiMo release", "excerpt": "I love the contrast with yesterday's open-source MiMo release, which put research chemistry (metal-organic frameworks stuff) front and center in the release notes. https://mimo.xiaomi.com/mimo-v2-6#co-scientist-for-materials-research", "context": "", "use_case": "research chemistry", "switched_from": "", "story": "Claude Opus 5.5"}, {"id": 796883650, "item_id": 49805228, "url": "https://news.ycombinator.com/item?id=49805228", "by": "vardalab", "date": "2026-09-22", "stance": "recommend", "quote": "flash models, DeepSeek, MiMo, GLM, I love those things", "excerpt": "Yeah, flash models, DeepSeek, MiMo, GLM, I love those things. For simple tasks like a daily routine shit, just setting up stuff and then doing the hard stuff in Claude/Codex, that's a reasonable approach for someone like me, a \"gentleman code farmer\", lol. And even lower tier stuff, I have the local models taking care of. Now that Jev is out I can finally have a true AI sysadmins managing my \"cloud in the basement\" h", "context": "homelab user", "use_case": "simple routine tasks", "switched_from": "", "story": "Claude Opus 5.5"}, {"id": 796779424, "item_id": 49798714, "url": "https://news.ycombinator.com/item?id=49798714", "by": "networked", "date": "2026-09-22", "stance": "neutral", "quote": "I asked MiMo-V2.6-Flash to fork and modify it in a straightforward way.", "excerpt": "I was curious to see how this would work with bzip2 and zstd. The source is public at https://github.com/nathanrs/gzipt, and I asked MiMo-V2.6-Flash to fork and modify it in a straightforward way. The answer is that bzip2 produces sequences that don't resemble human language: gzipt \\ --corpus data/tinyshakespeare.txt \\ --prompt $'MENENIUS:\\n' \\ --length 200 \\ ; MENENIUS: MtLUMSeptuttyyyxyxyxyxyvyyyxyxyxyxyvyyyxyxyxyx", "context": "", "use_case": "modifying code", "switched_from": "", "story": "Can gzip be a language model?"}, {"id": 796737056, "item_id": 49796066, "url": "https://news.ycombinator.com/item?id=49796066", "by": "nivance", "date": "2026-09-22", "stance": "recommend", "quote": "it sounds greate. I still have 50% of my quota this month, so I'll continue renewing to give it a try.", "excerpt": "it sounds greate. I still have 50% of my quota this month, so I'll continue renewing to give it a try.", "context": "", "use_case": "", "switched_from": "", "story": "MiMo v2.6"}, {"id": 796712656, "item_id": 49794541, "url": "https://news.ycombinator.com/item?id=49794541", "by": "wkcheng", "date": "2026-09-21", "stance": "neutral", "quote": "Are people using MiMo models as their daily drivers in a company setting?", "excerpt": "Are people using MiMo models as their daily drivers in a company setting? If so, how? I know they're available over OpenCode and directly from Xiaomi, but those are not great options. OpenCode Go straight up doesn't give any guarantees about training on your data, and Xiaomi says that they won't but it's unclear. With some models you can find hosting companies based in the EU or US, but then you don't know how they'r", "context": "", "use_case": "company daily driver model", "switched_from": "", "story": "MiMo v2.6"}, {"id": 796708912, "item_id": 49794307, "url": "https://news.ycombinator.com/item?id=49794307", "by": "GodelNumbering", "date": "2026-09-21", "stance": "recommend", "quote": "Mimo has been one of those models that I have been rooting for since the first I used it", "excerpt": "Mimo has been one of those models that I have been rooting for since the first I used it, the 2.5 pro which I have used quite a bit, was very concise, very aware of how much context needs to be read for which tasks and would always keep the context tight. Also surprisingly good at strategic thinking. I had published a comparison between it and Terra where Terra was found to be using much more avg context for similar ", "context": "", "use_case": "agentic tasks", "switched_from": "", "story": "MiMo v2.6"}, {"id": 796698192, "item_id": 49793637, "url": "https://news.ycombinator.com/item?id=49793637", "by": "volf_", "date": "2026-09-21", "stance": "neutral", "quote": "Averages ~25-35tok/s which isn't bad for a first attempt.", "excerpt": "I've got a working recipe to run this model on Dual DGX Spark: https://github.com/volfco/spark-vllm-docker/blob/main/recipes/mimo-v2.6-flash.yaml Averages ~25-35tok/s which isn't bad for a first attempt.", "context": "", "use_case": "running locally on Dual DGX Spark", "switched_from": "", "story": "MiMo v2.6"}, {"id": 796687968, "item_id": 49792998, "url": "https://news.ycombinator.com/item?id=49792998", "by": "rao-v", "date": "2026-09-21", "stance": "recommend", "quote": "I really like how transparent they\u2019ve been about the training of this model.", "excerpt": "I know we have strong views on what a truly open model is (open weights, open training data, open training code etc.) but I really like how transparent they\u2019ve been about the training of this model. The realtime dashboard they shared during training (https://mimo.xiaomi.com/rl/) was an incredible learning and teaching tool for me, and they\u2019ve been unusually comprehensive in sharing details about their methodology (ch", "context": "", "use_case": "open model", "switched_from": "", "story": "MiMo v2.6"}, {"id": 794771488, "item_id": 49673218, "url": "https://news.ycombinator.com/item?id=49673218", "by": "bellowsgulch", "date": "2026-09-12", "stance": "recommend", "quote": "mimo-v2.5-free, mimo-v2.5, deepseek-flash in that order", "excerpt": "mimo-v2.5-free, mimo-v2.5, deepseek-flash in that order, honestly don\u2019t even bother using qwen3.6-35b-a3b now unless i need uncensored tasks finished, mostly reverse engineering most engineering tasks don\u2019t require frontier llms when they get stuck, then i consider moving up to more capable models purchasing a claude plan seems widely unnecessary to me the tasks they do better than the average engineer cut both ways:", "context": "", "use_case": "engineering tasks", "switched_from": "", "story": "Ask HN: What default model do you use and why?"}, {"id": 793008145, "item_id": 49563009, "url": "https://news.ycombinator.com/item?id=49563009", "by": "benjiro29", "date": "2026-09-04", "stance": "warn", "quote": "MiMo did that a lot in the past", "excerpt": "3. The problem is also that at a lot of code bases are not designed around LLMs and their token usage. If you have a monolete codebase, you need to clearly define in the prompt what modules are involved. And even then, your wasting tokens with the first 1 or 2 steps where it needs to located the modules. if you have a github codebase, with a lot of your code into nice little repos, its even worse because then the mod", "context": "", "use_case": "", "switched_from": "", "story": "GPT-6 Astra"}, {"id": 791954242, "item_id": 49497140, "url": "https://news.ycombinator.com/item?id=49497140", "by": "sieve", "date": "2026-08-30", "stance": "compare", "quote": "MiMo is the dumbest of them all but good enough for basic stuff.", "excerpt": "I have used all three extensively. DS4 Flash is quite smart. I would rank MS 1.2 below it. MiMo is the dumbest of them all but good enough for basic stuff. MiMo wins handsomely if you want to think about your code for minutes at a time as you write. I use it to make changes as I think. I know it will screw up some stuff. I then switch to MS/DS4 once every few hours and have it do a code review and fix the broken stuf", "context": "", "use_case": "cheap code changes while thinking", "switched_from": "", "story": "Hy4 preview"}, {"id": 790838210, "item_id": 49427388, "url": "https://news.ycombinator.com/item?id=49427388", "by": "larsfaye", "date": "2026-08-24", "stance": "recommend", "quote": "Zed + OpenCode + Mimo/GLM/Kimi models are incredible. I haven't touched a \"frontier model\" for months.", "excerpt": "This is EXACTLY how I am working with these tools, you summarized everything about why I use this approach perfectly. Zed + OpenCode + Mimo/GLM/Kimi models are incredible. I haven't touched a \"frontier model\" for months. I really treat them as \"delegation layer\". They're systems that sit right alongside your stack, and at this point, at every level in it, like a connective tissue that intersects at all points. I can ", "context": "", "use_case": "AI coding", "switched_from": "", "story": "Coding expertise is going to collapse from AI reliance"}, {"id": 790318691, "item_id": 49394918, "url": "https://news.ycombinator.com/item?id=49394918", "by": "surgical_fire", "date": "2026-08-21", "stance": "recommend", "quote": "mostly GLM-5.3, DS-V4-Flash, and MiMo-2.5-pro", "excerpt": "Which is why I actually used all those models. At work I am stuck with Claude because it is what the employer provides. In my home setup I use a combination of different models; mostly GLM-5.3, DS-V4-Flash, and MiMo-2.5-pro. I vastly prefer my home setup.", "context": "", "use_case": "home setup", "switched_from": "", "story": "Bringing the cybersecurity capabilities of Claude Mythos 5 to more defenders"}, {"id": 786927155, "item_id": 49182947, "url": "https://news.ycombinator.com/item?id=49182947", "by": "surgical_fire", "date": "2026-08-05", "stance": "recommend", "quote": "DS and MiMo feel very snappy now", "excerpt": "I recently moved entirely to Pi, and I am really enjoying the journey so far. I am slowly shaping it to the way I like to work. It's a bit of a bumpy start, but there are two benefits I see from it: First is that you get to understand better what goes on behind the scenes. Since it is minimalistic, you have to think about your workflow, about what you want your agents to do, how you want them to do it, etc. It's a co", "context": "", "use_case": "coding model in Pi", "switched_from": "", "story": "Pi's Minimalism Is Its Advantage"}, {"id": 786792082, "item_id": 49174505, "url": "https://news.ycombinator.com/item?id=49174505", "by": "surgical_fire", "date": "2026-08-04", "stance": "recommend", "quote": "They are criminally cheap in a way that I don't care to spend tokens.", "excerpt": "> I've never touched the Chinese models (no particular reason) but it is having an impact as the frontier companies are pushing the price down as a defensive measure I warn you that it is addictive. I am using DS and MiMo on Pi.dev, and I don't see myself going back to OpenAI. They are criminally cheap in a way that I don't care to spend tokens. That leaves me space to experiment. And in terms of capability, well... ", "context": "uses Sonnet and Opus at work", "use_case": "coding via Pi", "switched_from": "", "story": "The AI Demand Bubble"}, {"id": 785791600, "item_id": 49111975, "url": "https://news.ycombinator.com/item?id=49111975", "by": "CapyToolkit", "date": "2026-07-30", "stance": "warn", "quote": "All behave dumb, don't follow instructions properly even when they write skills for themselves.", "excerpt": "Honestly, I have a similar impressions after testing every model except Sonnet, Opus or Claude. I always test them inside Claude vscode extension in the terminal, using the same skills or workflows from multiple skills. I tested Mimo V2.5 Pro, Qwen 3.8 Max Preview, GLM 5.2, Macaron V1 Venti and many other, smaller model whenever hype on X appears. All behave dumb, don't follow instructions properly even when they wri", "context": "tests many models", "use_case": "simple tasks", "switched_from": "", "story": "Kimi-code not performing well"}, {"id": 785312918, "item_id": 49082057, "url": "https://news.ycombinator.com/item?id=49082057", "by": "Frannky", "date": "2026-07-28", "stance": "recommend", "quote": "MiMo 2.5 Pro for reasoning", "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": "reasoning", "switched_from": "", "story": "\"Opus 5 is a really bad model\""}, {"id": 785288337, "item_id": 49080521, "url": "https://news.ycombinator.com/item?id=49080521", "by": "ignoramous", "date": "2026-07-28", "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": 783707200, "item_id": 48981700, "url": "https://news.ycombinator.com/item?id=48981700", "by": "dizhn", "date": "2026-07-20", "stance": "neutral", "quote": "the only currently really free offering is Mimo via their own CLI", "excerpt": "Yes I understood your other comment. I used free models for a long time. Less than 6 months ago there were a ton of options but the current free choices are very limited in number and usability. They run out way too quickly and some of them have smaller context sizes than paid offerings. As far as I know the only currently really free offering is Mimo via their own CLI.", "context": "", "use_case": "free model access", "switched_from": "", "story": "Stop Using OpenCode"}, {"id": 783407777, "item_id": 48962986, "url": "https://news.ycombinator.com/item?id=48962986", "by": "ignoramous", "date": "2026-07-18", "stance": "recommend", "quote": "Prefer the latter over DeepSeek v4 Pro because it costs the same while being equally good & less chattier for coding workloads.", "excerpt": "> It's cheap enough to where I don't think about cost, made even cheaper by DeepSeek having the most effective and cheap caching in the industry I use DeepSeek v4 Flash & MiMo v2.5 Pro. Prefer the latter over DeepSeek v4 Pro because it costs the same while being equally good & less chattier for coding workloads. Although, I've begun experimenting with Hy3 (as an in-between Flash & Pro) & GLM 5.2 (for long-horizon tas", "context": "", "use_case": "coding workloads", "switched_from": "", "story": "The Kimi K3 Moment"}, {"id": 783005153, "item_id": 48937822, "url": "https://news.ycombinator.com/item?id=48937822", "by": "ignoramous", "date": "2026-07-16", "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": 780471573, "item_id": 48779473, "url": "https://news.ycombinator.com/item?id=48779473", "by": "SwellJoe", "date": "2026-07-03", "stance": "neutral", "quote": "MiMo has the cheapest tokens for a 1T+ model in the game", "excerpt": "I recently wrote up how I run local LLMs, because several folks had asked (https://swelljoe.com/post/how-i-run-local-llms/) and I think even my setup, which I spent maybe $4200 on, half on a Strix Halo and half on upgrades for my desktop, would be too expensive to justify today. I bought before prices went through the roof, and only did so because I like to tinker with hardware...not because I expected it to ever pay", "context": "", "use_case": "cheap API tokens", "switched_from": "", "story": "Jamesob's guide to running SOTA LLMs locally"}, {"id": 780015572, "item_id": 48750973, "url": "https://news.ycombinator.com/item?id=48750973", "by": "SwellJoe", "date": "2026-07-01", "stance": "neutral", "quote": "MiMo has token plans that are a pretty good deal (though confusing", "excerpt": "North Mini Code is on par with the similar-sized MoE Qwen 3.6 and Gemma 4. It benches a little better than Gemma 4 26b a4b and a little worse than Qwen 3.6 35b a3b for most coding-related and agentic tasks. For my use case, or at least the use case I've tested, it is a little worse than both for surfacing security bugs, but all of the MoE models at this size are pretty bad at finding security bugs (they hallucinate a", "context": "", "use_case": "token plans", "switched_from": "", "story": "Qwen 3.6 27B is the sweet spot for local development"}, {"id": 779940115, "item_id": 48746257, "url": "https://news.ycombinator.com/item?id=48746257", "by": "coder543", "date": "2026-07-01", "stance": "neutral", "quote": "I was doing it just fine on llama-server with 200k+ of context.", "excerpt": "It\u2019s not FUD. It is my actual, lived experience. FUD is false, which this is not. I use both vLLM and llama-server. vLLM is very painful, even with the Spark community docker image. It is slow to start, it does not support 3-bit dynamic quants well, and it takes a lot of tweaking to get it to run well for each model I want to try out, which is made worse by the slow starts. I\u2019m glad you\u2019ve had a better experience? I ", "context": "", "use_case": "local model", "switched_from": "", "story": "Qwen 3.6 27B is the sweet spot for local development"}, {"id": 778794865, "item_id": 48674679, "url": "https://news.ycombinator.com/item?id=48674679", "by": "bel8", "date": "2026-06-25", "stance": "recommend", "quote": "You'll get 10x more value by using MiMo 2.5 or DeepSeek Flash which is what I use for most code.", "excerpt": "GLM 5.2 is priced at 1/3 of Opus. Depending on who you ask for a napkin math. Not worth it in my opinion. You'll get 10x more value by using MiMo 2.5 or DeepSeek Flash which is what I use for most code.", "context": "", "use_case": "coding", "switched_from": "", "story": "Elevated error rate across multiple models"}, {"id": 776119664, "item_id": 48507479, "url": "https://news.ycombinator.com/item?id=48507479", "by": "mdasen", "date": "2026-06-12", "stance": "recommend", "quote": "cached inputs are practically free", "excerpt": "I find that I don't use a ton of output tokens. I'm usually around 95% cached input, 4% input, and 1% output. For me, the big thing with MiMo-V2.5-Pro and DeepSeek V4-Pro is that cached inputs are practically free. Kimi K2.7 Code is 53x more expensive for cached inputs which is 95% of my costs. If I use 95M cached input tokens, 4M input tokens, and 1M output tokens, that'd be: $18 for cached input on Kimi K2.7 Code v", "context": "heavy cached-input API user", "use_case": "AI coding", "switched_from": "", "story": "Kimi K2.7-Code: open-source coding model with better token efficiency"}]}