{"slug": "claude-haiku", "name": "Claude Haiku", "category": "ai", "category_name": "AI models and assistants", "description": "Anthropic's smallest, fastest and cheapest Claude large language model.", "recommenders": 30, "commenters": 58, "warners": 5, "score": 27.521, "rising": 11.42, "first": "2025-11-09", "latest": "2026-10-29", "why": "Recommended as a cheap, fast model for subagents and simple, clearly specified tasks, and for turning natural language into structured JSON.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 31, "neutral": 19, "compare": 6, "warn": 6}, "spark": [2, 2, 2, 2, 3, 0, 0, 1, 2, 3, 5, 9], "listed": true, "loved": ["Subagent workhorse", "Cheap and fast"], "caveat": "Complex coding", "short_quote": "We use Haiku 4.5 to turn users' natural language queries and requests into fairly complex structured specs for interior design and construction. It has near perfect accuracy.", "icon": true, "domain": "haiku-os.org", "github_repo": null, "official_url": "https://www.anthropic.com/claude/haiku", "pricing": "paid", "aliases": ["Haiku", "Claude Haiku", "Claude Haiku 4.5", "Haiku 5.5", "Claude Haiku 5.5", "Haiku 4.5", "claude-haiku-4.5"], "verdict": [{"text": "Recommended as a cheap, fast model for subagents and simple, clearly specified tasks", "cites": [{"id": 802679058, "item_id": 50167441, "url": "https://news.ycombinator.com/item?id=50167441", "by": "A_D_E_P_T", "date": "2026-10-20", "stance": "recommend", "quote": "Haiku 5.5 is decent, cheap, and fast.", "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": "grunt work coding", "switched_from": "", "story": "Ask HN: Best value/money AI coding plan bundle?"}, {"id": 785336722, "item_id": 49083545, "url": "https://news.ycombinator.com/item?id=49083545", "by": "oggreen", "date": "2026-07-28", "stance": "recommend", "quote": "at work Haiku almost always works just as well as Sonnet does and its much cheaper/faster", "excerpt": "I use Sonnet 80% of the time when I'm on claude.ai and on Claude Code my main agent is Opus and typically my sub-agents are Sonnet. I use this setup mainly because I genuinely don't know if anything I'm doing is complex enough to need Fable and I don't want to run up my usage too much. It's basically like trying to kill a gnat with a sledgehammer. Hell, at work Haiku almost always works just as well as Sonnet does an", "context": "", "use_case": "work tasks", "switched_from": "", "story": "Why I prefer Opus 5 to Fable 5"}, {"id": 799947376, "item_id": 49996711, "url": "https://news.ycombinator.com/item?id=49996711", "by": "mariocesar", "date": "2026-10-07", "stance": "recommend", "quote": "I use haiku for things that needs to be quick, have really clear instructions.", "excerpt": "I have a zsh functions that calls claude code with haiku to suggest commit messages, is faster and the instructions are two lines. I also have an \"ask\" script that I use daily to ask simple stuff, it can access websearch and webfetch, it's more than enough to parse logs, ask for commands, quick research on the internet, small stuff. https://github.com/mariocesar/dotfiles/blob/main/common/.local/bin/ask I use haiku fo", "context": "", "use_case": "commit messages, quick queries", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 799991168, "item_id": 49999448, "url": "https://news.ycombinator.com/item?id=49999448", "by": "jeremyjh", "date": "2026-10-07", "stance": "recommend", "quote": "I would mostly use Haiku in task or explorer subagents.", "excerpt": "If you can't get any coding done with 100K context that is either a broken model, a broken harness or a skill issue. I would mostly use Haiku in task or explorer subagents. I'm not saying I stay under that on every task, but I do have quite a few sessions that cap out well below that, so that price difference would be very meaningful. I use Luna for this day in and out and its excellent - if Haiku is that much better", "context": "", "use_case": "subagents", "switched_from": "Luna", "story": "Claude Haiku 5.5"}, {"id": 793502833, "item_id": 49593927, "url": "https://news.ycombinator.com/item?id=49593927", "by": "dxjxjdjsssb", "date": "2026-09-07", "stance": "recommend", "quote": "a skill/script combo suitable for Haiku to accomplish the task, and now that task happens at minimal cost", "excerpt": "I use skills for offloading work onto subagents. By configuring the skill to use a specific model it gets enforced at the harness instead of depending on the good will of the orchestrating model to actually delegate. This also saves context. Today Fable had to fetch a zip file from a web page with a eula prompt, then get at a file in a disk image in the zip. This is something that will need to happen a lot as part of", "context": "", "use_case": "cheap subagent tasks", "switched_from": "", "story": "Ask HN: How do you manage skills files?"}]}, {"text": ", and for turning natural language into structured JSON.", "cites": [{"id": 799967856, "item_id": 49997991, "url": "https://news.ycombinator.com/item?id=49997991", "by": "enraged_camel", "date": "2026-10-07", "stance": "recommend", "quote": "We use Haiku 4.5 to turn users' natural language queries and requests into fairly complex structured specs for interior design and construction. It has near perfect accuracy.", "excerpt": "Not really. We use Haiku 4.5 to turn users' natural language queries and requests into fairly complex structured specs for interior design and construction. It has near perfect accuracy.", "context": "builds interior design/construction product", "use_case": "natural language to structured specs", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 796875744, "item_id": 49804734, "url": "https://news.ycombinator.com/item?id=49804734", "by": "enraged_camel", "date": "2026-09-22", "stance": "recommend", "quote": "It continues to be absurdly capable for converting natural language to structured JSON", "excerpt": "We use Haiku 4.5 inside our product. It continues to be absurdly capable for converting natural language to structured JSON based on a set of fairly complex business rules.", "context": "uses it inside their product", "use_case": "natural language to structured JSON", "switched_from": "", "story": "Claude Opus 5.5"}, {"id": 734894928, "item_id": 45930933, "url": "https://news.ycombinator.com/item?id=45930933", "by": "barefootford", "date": "2025-11-14", "stance": "recommend", "quote": "I switched from structured outputs on OpenAI apis to unstructured on Claude (haiku 4.5) and haven't had any issues (yet)", "excerpt": "I switched from structured outputs on OpenAI apis to unstructured on Claude (haiku 4.5) and haven't had any issues (yet). But guarantees are always nice.", "context": "", "use_case": "structured outputs via API", "switched_from": "OpenAI API", "story": "Structured outputs on the Claude Developer Platform"}]}, {"text": " The most common complaint is that it falls short on serious or complex coding work", "cites": [{"id": 774009587, "item_id": 48375599, "url": "https://news.ycombinator.com/item?id=48375599", "by": "bel8", "date": "2026-06-02", "stance": "warn", "quote": "They are cute but for serious coding they tend to waste your expensive time.", "excerpt": "It's a start and I welcome competition but I don't think I ever used small cloud models like Haiku 4.5. They are cute but for serious coding they tend to waste your expensive time. And this certainly wont bring me back to GitHub Copilot which I cancelled yesterday. GitHub Copilot had competitive pricing until yesterday when they changed from per-request to one of the most expensive per-token quotas. Seriously, take a", "context": "", "use_case": "coding", "switched_from": "", "story": "Microsoft's MAI-Code-1-Flash Scores 51% SWE-Bench Pro with Just 5B Active Params"}, {"id": 761391554, "item_id": 47586972, "url": "https://news.ycombinator.com/item?id=47586972", "by": "giancarlostoro", "date": "2026-03-31", "stance": "warn", "quote": "when I use them via JetBrains AI it feels like only Opus is good enough", "excerpt": "I'm guessing their newer models are taking way more compute than they can afford to give away. The biggest challenge of AI will eventually be, how to bring down how much compute a powerful model takes. I hope Claude puts more emphasis into making Haiku and Sonnet better, when I use them via JetBrains AI it feels like only Opus is good enough, for whatever odd reason.", "context": "", "use_case": "", "switched_from": "", "story": "Anthropic: Claude Code users hitting usage limits 'way faster than expected'"}, {"id": 799944640, "item_id": 49996540, "url": "https://news.ycombinator.com/item?id=49996540", "by": "giancarlostoro", "date": "2026-10-07", "stance": "neutral", "quote": "any time I ask Opus to \"try\" letting Haiku do the work, it just falls apart", "excerpt": "I with they'd give Haiku like 400k tokens roughly, I think between 400k or even 600k tokens is a sweet spot, but Haiku is basically designed to be for small edits is my understanding, but it sucks because any time I ask Opus to \"try\" letting Haiku do the work, it just falls apart and Opus comes back and tells me it switched to Sonnet (even before Sonnet finally jumped up to 5.x). I will try the new Haiku, but it woul", "context": "", "use_case": "file editing for Opus/Sonnet", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 797400368, "item_id": 49837523, "url": "https://news.ycombinator.com/item?id=49837523", "by": "sznio", "date": "2026-09-24", "stance": "warn", "quote": "current Haiku is useless", "excerpt": "Nothing, because current Haiku is useless. But I use Qwen3.6 35B-A3B and it's both much faster and much more capable than Haiku. And Qwen3.8 27B can do so much more, decently replaces Sonnet, albeit not as fast. Luna doesn't feel more capable nor faster than Qwen3.8. I think Anthropic could surprise us with some more innovation in this niche.", "context": "", "use_case": "", "switched_from": "", "story": "Claude Opus 5.5"}, {"id": 750296226, "item_id": 46893514, "url": "https://news.ycombinator.com/item?id=46893514", "by": "IgorPartola", "date": "2026-02-04", "stance": "warn", "quote": "In practice the subagents took so many steps and wrote so much documentation that it became not worth it.", "excerpt": "So I have gotten pretty good at managing context such that my $20 Claude subscription rarely runs out of its quota but I still do hit it sometimes. I use Sonnet 99% of the time. Mostly this comes down to giving it specific task and using /clear frequently. I also ask it to update its own notes frequently so it doesn\u2019t have to explore the whole codebase as often. But I was really disappointed when I tried to use subag", "context": "$20 subscriber", "use_case": "subagents for routine tasks", "switched_from": "", "story": "Claude Code: connect to a local model when your quota runs out"}]}, {"text": ", and some say newer versions got slower.", "cites": [{"id": 799958257, "item_id": 49997391, "url": "https://news.ycombinator.com/item?id=49997391", "by": "d1l", "date": "2026-10-07", "stance": "warn", "quote": "Results so far are trash. Prompt leakage even. And it\u2019s slower.", "excerpt": "At work we use haiku 4.5 for a handful of latency sensitive tasks that are fairly simple. It performs well. Just started testing 5.5 as I\u2019ve been anticipating a nice improvement since it was teased. Results so far are trash. Prompt leakage even. And it\u2019s slower. I guess it\u2019s cheap but I think they got the balance wrong on this.", "context": "", "use_case": "latency-sensitive simple tasks", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 797244400, "item_id": 49827775, "url": "https://news.ycombinator.com/item?id=49827775", "by": "RugnirViking", "date": "2026-09-24", "stance": "neutral", "quote": "I used to use haiku for the speed. Now its just as slow as the rest.", "excerpt": "> Nobody needs to code at 600 words per second. I do. I used to use haiku for the speed. Now its just as slow as the rest. Speed is my #1 ranking of how good a model is", "context": "", "use_case": "fast coding", "switched_from": "", "story": "Mercury 2.5 LLM hits 770 tokens per second"}]}], "praise": [{"label": "Subagent workhorse", "summary": "Often paired with Opus or Sonnet as the cheap model for delegated grunt work, tests and explorer subagents.", "quotes": [{"id": 802679058, "item_id": 50167441, "url": "https://news.ycombinator.com/item?id=50167441", "by": "A_D_E_P_T", "date": "2026-10-20", "stance": "recommend", "quote": "Haiku 5.5 is decent, cheap, and fast.", "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": "grunt work coding", "switched_from": "", "story": "Ask HN: Best value/money AI coding plan bundle?"}, {"id": 799991168, "item_id": 49999448, "url": "https://news.ycombinator.com/item?id=49999448", "by": "jeremyjh", "date": "2026-10-07", "stance": "recommend", "quote": "I would mostly use Haiku in task or explorer subagents.", "excerpt": "If you can't get any coding done with 100K context that is either a broken model, a broken harness or a skill issue. I would mostly use Haiku in task or explorer subagents. I'm not saying I stay under that on every task, but I do have quite a few sessions that cap out well below that, so that price difference would be very meaningful. I use Luna for this day in and out and its excellent - if Haiku is that much better", "context": "", "use_case": "subagents", "switched_from": "Luna", "story": "Claude Haiku 5.5"}, {"id": 798130226, "item_id": 49883139, "url": "https://news.ycombinator.com/item?id=49883139", "by": "sajithdilshan", "date": "2026-09-28", "stance": "neutral", "quote": "Haiku for running tests", "excerpt": "I use Claude Code everyday for work and the main model I use is Opus (For planning, breaking down tasks, writing tickets, implementation, etc.) and Haiku for running tests. Honestly have no idea what is the use case for Sonnet", "context": "", "use_case": "running tests", "switched_from": "", "story": "Sonnet 5.5"}]}, {"label": "Cheap and fast", "summary": "Described as cheap and fast, and as good as Sonnet for many simple work tasks at lower cost.", "quotes": [{"id": 802679058, "item_id": 50167441, "url": "https://news.ycombinator.com/item?id=50167441", "by": "A_D_E_P_T", "date": "2026-10-20", "stance": "recommend", "quote": "Haiku 5.5 is decent, cheap, and fast.", "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": "grunt work coding", "switched_from": "", "story": "Ask HN: Best value/money AI coding plan bundle?"}, {"id": 785336722, "item_id": 49083545, "url": "https://news.ycombinator.com/item?id=49083545", "by": "oggreen", "date": "2026-07-28", "stance": "recommend", "quote": "at work Haiku almost always works just as well as Sonnet does and its much cheaper/faster", "excerpt": "I use Sonnet 80% of the time when I'm on claude.ai and on Claude Code my main agent is Opus and typically my sub-agents are Sonnet. I use this setup mainly because I genuinely don't know if anything I'm doing is complex enough to need Fable and I don't want to run up my usage too much. It's basically like trying to kill a gnat with a sledgehammer. Hell, at work Haiku almost always works just as well as Sonnet does an", "context": "", "use_case": "work tasks", "switched_from": "", "story": "Why I prefer Opus 5 to Fable 5"}, {"id": 749760897, "item_id": 46860056, "url": "https://news.ycombinator.com/item?id=46860056", "by": "Soerensen", "date": "2026-02-02", "stance": "recommend", "quote": "Tweaks and refinements use Haiku. Users can't tell the difference for small edits, and it cut iteration costs ~80%.", "excerpt": "Our biggest cost multiplier was \"conversational drift\" - not the initial call, but what happens when you let users iterate. In our email marketing tool, a user might say \"make it more punchy\" \u2192 AI rewrites \u2192 \"actually, more professional\" \u2192 rewrite \u2192 \"can we A/B test both versions?\" \u2192 now you're generating multiple variants. One \"simple\" email could spiral into 15+ LLM calls. What worked for us: 1. *Session-level budg", "context": "builds an email marketing tool", "use_case": "cheap iteration edits", "switched_from": "", "story": "Ask HN: What's your biggest LLM cost multiplier?"}]}, {"label": "Simple tasks", "summary": "Used for quick, well-defined jobs like commit messages, log parsing, tagging and latency-sensitive tasks.", "quotes": [{"id": 799947376, "item_id": 49996711, "url": "https://news.ycombinator.com/item?id=49996711", "by": "mariocesar", "date": "2026-10-07", "stance": "recommend", "quote": "I use haiku for things that needs to be quick, have really clear instructions.", "excerpt": "I have a zsh functions that calls claude code with haiku to suggest commit messages, is faster and the instructions are two lines. I also have an \"ask\" script that I use daily to ask simple stuff, it can access websearch and webfetch, it's more than enough to parse logs, ask for commands, quick research on the internet, small stuff. https://github.com/mariocesar/dotfiles/blob/main/common/.local/bin/ask I use haiku fo", "context": "", "use_case": "commit messages, quick queries", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 799958256, "item_id": 49997391, "url": "https://news.ycombinator.com/item?id=49997391", "by": "d1l", "date": "2026-10-07", "stance": "recommend", "quote": "At work we use haiku 4.5 for a handful of latency sensitive tasks that are fairly simple. It performs well.", "excerpt": "At work we use haiku 4.5 for a handful of latency sensitive tasks that are fairly simple. It performs well. Just started testing 5.5 as I\u2019ve been anticipating a nice improvement since it was teased. Results so far are trash. Prompt leakage even. And it\u2019s slower. I guess it\u2019s cheap but I think they got the balance wrong on this.", "context": "", "use_case": "latency-sensitive simple tasks", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 804546960, "item_id": 50284185, "url": "https://news.ycombinator.com/item?id=50284185", "by": "cj00", "date": "2026-10-29", "stance": "neutral", "quote": "then I use Haiku to do some categorization and auto tagging", "excerpt": "OCR on the thumbnail is sometimes good enough for me since I\u2019ll remember the teaser blurb that they put in the image but leave out of the title. So the LLM description service can handle that and maybe it\u2019ll add in \u201ctwo people standing in front of a boat\u201d and something will hit in a search. But I\u2019ve actually started pulling in more than just the thumbnail. With the YouTube API I\u2019m pulling in the description and all t", "context": "builds a semantic search project", "use_case": "categorizing and tagging video metadata", "switched_from": "", "story": "Semantic search across major museums"}]}, {"label": "Structured output", "summary": "Reported as highly accurate at converting natural language into complex structured specs and JSON.", "quotes": [{"id": 799967856, "item_id": 49997991, "url": "https://news.ycombinator.com/item?id=49997991", "by": "enraged_camel", "date": "2026-10-07", "stance": "recommend", "quote": "We use Haiku 4.5 to turn users' natural language queries and requests into fairly complex structured specs for interior design and construction. It has near perfect accuracy.", "excerpt": "Not really. We use Haiku 4.5 to turn users' natural language queries and requests into fairly complex structured specs for interior design and construction. It has near perfect accuracy.", "context": "builds interior design/construction product", "use_case": "natural language to structured specs", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 796875744, "item_id": 49804734, "url": "https://news.ycombinator.com/item?id=49804734", "by": "enraged_camel", "date": "2026-09-22", "stance": "recommend", "quote": "It continues to be absurdly capable for converting natural language to structured JSON", "excerpt": "We use Haiku 4.5 inside our product. It continues to be absurdly capable for converting natural language to structured JSON based on a set of fairly complex business rules.", "context": "uses it inside their product", "use_case": "natural language to structured JSON", "switched_from": "", "story": "Claude Opus 5.5"}, {"id": 734894928, "item_id": 45930933, "url": "https://news.ycombinator.com/item?id=45930933", "by": "barefootford", "date": "2025-11-14", "stance": "recommend", "quote": "I switched from structured outputs on OpenAI apis to unstructured on Claude (haiku 4.5) and haven't had any issues (yet)", "excerpt": "I switched from structured outputs on OpenAI apis to unstructured on Claude (haiku 4.5) and haven't had any issues (yet). But guarantees are always nice.", "context": "", "use_case": "structured outputs via API", "switched_from": "OpenAI API", "story": "Structured outputs on the Claude Developer Platform"}]}, {"label": "Surprisingly capable", "summary": "Some users were surprised by how well it handled harder work and say it keeps getting smarter.", "quotes": [{"id": 740311520, "item_id": 46269470, "url": "https://news.ycombinator.com/item?id=46269470", "by": "llmthrow0827", "date": "2025-12-14", "stance": "recommend", "quote": "I used Haiku with Claude Code during the outage, and was surprised at how well it did.", "excerpt": "I used Haiku with Claude Code during the outage, and was surprised at how well it did. I'm going to try mixing it in more to save usage credits.", "context": "", "use_case": "cheaper coding model in Claude Code", "switched_from": "", "story": "Elevated errors across many models"}, {"id": 800975536, "item_id": 50060971, "url": "https://news.ycombinator.com/item?id=50060971", "by": "JMKH42", "date": "2026-10-12", "stance": "recommend", "quote": "Now even Haiku at low effort does.", "excerpt": "Using the paid anthrotpic models daily, it has not been my experience that they are just fixing public examples of foolishness, but are generally getting smarter. For example I have my own private, obscure bike racing physics test I use that the newer low tier models pass now. Before fable, the top tier models didn't pass it. Now even Haiku at low effort does.", "context": "uses paid Anthropic models daily", "use_case": "", "switched_from": "", "story": "The gap in shared understanding of LLM capability is widening"}, {"id": 749162145, "item_id": 46822634, "url": "https://news.ycombinator.com/item?id=46822634", "by": "freeqaz", "date": "2026-01-30", "stance": "recommend", "quote": "just letting Haiku rip. Honestly, it blows me away.", "excerpt": "I've been working on decompiling Dance Central 3 with AI and it's been insane. It's an Xbox 360 game that leverages the Kinect to track your body as your dance. It's a great game, but even with an emulator, it's still dependent on the Kinect hardware which is proprietary and has limited supply. Fortunately, a Debug build of this game was found on a dev unit (somehow), and that build does _not_ have crazy optimization", "context": "", "use_case": "decompiling code", "switched_from": "", "story": "PlayStation 2 Recompilation Project Is Absolutely Incredible"}]}], "caveats": [{"label": "Complex coding", "summary": "Several users say it struggles with serious coding and falls apart when given larger tasks.", "quotes": [{"id": 774009587, "item_id": 48375599, "url": "https://news.ycombinator.com/item?id=48375599", "by": "bel8", "date": "2026-06-02", "stance": "warn", "quote": "They are cute but for serious coding they tend to waste your expensive time.", "excerpt": "It's a start and I welcome competition but I don't think I ever used small cloud models like Haiku 4.5. They are cute but for serious coding they tend to waste your expensive time. And this certainly wont bring me back to GitHub Copilot which I cancelled yesterday. GitHub Copilot had competitive pricing until yesterday when they changed from per-request to one of the most expensive per-token quotas. Seriously, take a", "context": "", "use_case": "coding", "switched_from": "", "story": "Microsoft's MAI-Code-1-Flash Scores 51% SWE-Bench Pro with Just 5B Active Params"}, {"id": 799944640, "item_id": 49996540, "url": "https://news.ycombinator.com/item?id=49996540", "by": "giancarlostoro", "date": "2026-10-07", "stance": "neutral", "quote": "any time I ask Opus to \"try\" letting Haiku do the work, it just falls apart", "excerpt": "I with they'd give Haiku like 400k tokens roughly, I think between 400k or even 600k tokens is a sweet spot, but Haiku is basically designed to be for small edits is my understanding, but it sucks because any time I ask Opus to \"try\" letting Haiku do the work, it just falls apart and Opus comes back and tells me it switched to Sonnet (even before Sonnet finally jumped up to 5.x). I will try the new Haiku, but it woul", "context": "", "use_case": "file editing for Opus/Sonnet", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 761391554, "item_id": 47586972, "url": "https://news.ycombinator.com/item?id=47586972", "by": "giancarlostoro", "date": "2026-03-31", "stance": "warn", "quote": "when I use them via JetBrains AI it feels like only Opus is good enough", "excerpt": "I'm guessing their newer models are taking way more compute than they can afford to give away. The biggest challenge of AI will eventually be, how to bring down how much compute a powerful model takes. I hope Claude puts more emphasis into making Haiku and Sonnet better, when I use them via JetBrains AI it feels like only Opus is good enough, for whatever odd reason.", "context": "", "use_case": "", "switched_from": "", "story": "Anthropic: Claude Code users hitting usage limits 'way faster than expected'"}]}, {"label": "Newer versions slower", "summary": "Some found the 5.5 release slower and worse than before, including prompt leakage.", "quotes": [{"id": 799958257, "item_id": 49997391, "url": "https://news.ycombinator.com/item?id=49997391", "by": "d1l", "date": "2026-10-07", "stance": "warn", "quote": "Results so far are trash. Prompt leakage even. And it\u2019s slower.", "excerpt": "At work we use haiku 4.5 for a handful of latency sensitive tasks that are fairly simple. It performs well. Just started testing 5.5 as I\u2019ve been anticipating a nice improvement since it was teased. Results so far are trash. Prompt leakage even. And it\u2019s slower. I guess it\u2019s cheap but I think they got the balance wrong on this.", "context": "", "use_case": "latency-sensitive simple tasks", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 797244400, "item_id": 49827775, "url": "https://news.ycombinator.com/item?id=49827775", "by": "RugnirViking", "date": "2026-09-24", "stance": "neutral", "quote": "I used to use haiku for the speed. Now its just as slow as the rest.", "excerpt": "> Nobody needs to code at 600 words per second. I do. I used to use haiku for the speed. Now its just as slow as the rest. Speed is my #1 ranking of how good a model is", "context": "", "use_case": "fast coding", "switched_from": "", "story": "Mercury 2.5 LLM hits 770 tokens per second"}]}, {"label": "Price/performance", "summary": "Some see cheaper competitors like DeepSeek or local Qwen models as better value.", "quotes": [{"id": 800204071, "item_id": 50012754, "url": "https://news.ycombinator.com/item?id=50012754", "by": "gregwebs", "date": "2026-10-08", "stance": "warn", "quote": "Haiku also doesn't seem like a competitive price/performance mix.", "excerpt": "I have been using DeepSeek 4.1 flash intensively for over a month. If I run it all day long it costs $1-2. Its fast. Previously I was always quickly running up to my Claude/Codex 5 hour window (on the $20/month plan). The cost savings of DeepSeek is real as shown in this article and I am using subsidized plans. DeepSeek is horrible at grilling sessions (the /grill* skills to make technical decisions). It doesn't know", "context": "", "use_case": "", "switched_from": "", "story": "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"}, {"id": 797400368, "item_id": 49837523, "url": "https://news.ycombinator.com/item?id=49837523", "by": "sznio", "date": "2026-09-24", "stance": "warn", "quote": "current Haiku is useless", "excerpt": "Nothing, because current Haiku is useless. But I use Qwen3.6 35B-A3B and it's both much faster and much more capable than Haiku. And Qwen3.8 27B can do so much more, decently replaces Sonnet, albeit not as fast. Luna doesn't feel more capable nor faster than Qwen3.8. I think Anthropic could surprise us with some more innovation in this niche.", "context": "", "use_case": "", "switched_from": "", "story": "Claude Opus 5.5"}, {"id": 797228289, "item_id": 49826768, "url": "https://news.ycombinator.com/item?id=49826768", "by": "anshumankmr", "date": "2026-09-24", "stance": "neutral", "quote": "Absolute fucking L from anthropic for not introducing a newer cheaper Haiku4.5.", "excerpt": "Absolute fucking L from anthropic for not introducing a newer cheaper Haiku4.5. We have switched to GPT 5.6 luna (for some simpler workloads) and will switch to this too.", "context": "", "use_case": "", "switched_from": "", "story": "GPT-6 Sol and Luna"}]}, {"label": "Subagent overhead", "summary": "One user found Haiku subagents took too many steps and wrote too much documentation to be worth it.", "quotes": [{"id": 750296226, "item_id": 46893514, "url": "https://news.ycombinator.com/item?id=46893514", "by": "IgorPartola", "date": "2026-02-04", "stance": "warn", "quote": "In practice the subagents took so many steps and wrote so much documentation that it became not worth it.", "excerpt": "So I have gotten pretty good at managing context such that my $20 Claude subscription rarely runs out of its quota but I still do hit it sometimes. I use Sonnet 99% of the time. Mostly this comes down to giving it specific task and using /clear frequently. I also ask it to update its own notes frequently so it doesn\u2019t have to explore the whole codebase as often. But I was really disappointed when I tried to use subag", "context": "$20 subscriber", "use_case": "subagents for routine tasks", "switched_from": "", "story": "Claude Code: connect to a local model when your quota runs out"}]}, {"label": "Complex structures", "summary": "One user says it struggles with complex structured outputs.", "quotes": [{"id": 785336722, "item_id": 49083545, "url": "https://news.ycombinator.com/item?id=49083545", "by": "oggreen", "date": "2026-07-28", "stance": "recommend", "quote": "at work Haiku almost always works just as well as Sonnet does and its much cheaper/faster", "excerpt": "I use Sonnet 80% of the time when I'm on claude.ai and on Claude Code my main agent is Opus and typically my sub-agents are Sonnet. I use this setup mainly because I genuinely don't know if anything I'm doing is complex enough to need Fable and I don't want to run up my usage too much. It's basically like trying to kill a gnat with a sledgehammer. Hell, at work Haiku almost always works just as well as Sonnet does an", "context": "", "use_case": "work tasks", "switched_from": "", "story": "Why I prefer Opus 5 to Fable 5"}]}], "who": [{"by": "cj00", "context": "builds a semantic search project", "url": "https://news.ycombinator.com/item?id=50284185", "date": "2026-10-29"}, {"by": "JMKH42", "context": "uses paid Anthropic models daily", "url": "https://news.ycombinator.com/item?id=50060971", "date": "2026-10-12"}, {"by": "enraged_camel", "context": "builds interior design/construction product", "url": "https://news.ycombinator.com/item?id=49997991", "date": "2026-10-07"}, {"by": "lproven", "context": "owns ThinkPad X200/X220 machines", "url": "https://news.ycombinator.com/item?id=49896293", "date": "2026-09-29"}, {"by": "ethin", "context": "cares about accessibility, likely screen reader user", "url": "https://news.ycombinator.com/item?id=49503262", "date": "2026-08-30"}, {"by": "jmmv", "context": "has an old Powerbook G4", "url": "https://news.ycombinator.com/item?id=49502240", "date": "2026-08-30"}, {"by": "carpo", "context": "built a local Electron video indexing app", "url": "https://news.ycombinator.com/item?id=48231191", "date": "2026-05-21"}, {"by": "SyneRyder", "context": "occasional Haiku user", "url": "https://news.ycombinator.com/item?id=48124892", "date": "2026-05-13"}, {"by": "shevy-java", "context": "used it many years ago", "url": "https://news.ycombinator.com/item?id=48124763", "date": "2026-05-13"}, {"by": "heraldgeezer", "context": "free tier user", "url": "https://news.ycombinator.com/item?id=46902475", "date": "2026-02-05"}, {"by": "Soerensen", "context": "builds an email marketing tool", "url": "https://news.ycombinator.com/item?id=46860056", "date": "2026-02-02"}, {"by": "Froedlich", "context": "mostly retired", "url": "https://news.ycombinator.com/item?id=46332192", "date": "2025-12-19"}], "tags": ["for coding subagents", "for structured extraction", "for latency-sensitive tasks", "for classification and tagging"], "use_cases": ["coding", "daily driver os", "subagents", "latency-sensitive simple tasks", "desktop os", "categorizing and tagging video metadata"], "compared": [{"slug": "claude-opus", "name": "Claude Opus", "co": 24, "switched_from": 0, "switched_to": 0}, {"slug": "claude-sonnet", "name": "Claude Sonnet", "co": 21, "switched_from": 0, "switched_to": 1}, {"slug": "gpt-luna", "name": "GPT Luna", "co": 11, "switched_from": 1, "switched_to": 1}, {"slug": "claude-code", "name": "Claude Code", "co": 10, "switched_from": 0, "switched_to": 0}, {"slug": "deepseek-v4-flash", "name": "DeepSeek V4 Flash", "co": 8, "switched_from": 0, "switched_to": 0}, {"slug": "claude-fable", "name": "Claude Fable", "co": 6, "switched_from": 0, "switched_to": 0}, {"slug": "qwen-3-6", "name": "Qwen 3.6", "co": 4, "switched_from": 0, "switched_to": 1}, {"slug": "codex", "name": "Codex", "co": 4, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=49996437", "title": "Claude Haiku 5.5", "mentions": 9}, {"url": "https://news.ycombinator.com/item?id=49499867", "title": "Haiku R1/beta6 has been released", "mentions": 5}, {"url": "https://news.ycombinator.com/item?id=49803892", "title": "Claude Opus 5.5", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=49672966", "title": "Ask HN: What default model do you use and why?", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=48124002", "title": "Haiku", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=50194522", "title": "Semantic search across major museums", "mentions": 1}], "sentiment_by_month": {"2025-11": {"recommend": 2, "compare": 1}, "2025-12": {"recommend": 2, "neutral": 1}, "2026-01": {"neutral": 1, "recommend": 2}, "2026-02": {"recommend": 2, "warn": 1, "neutral": 2}, "2026-03": {"recommend": 3, "warn": 1, "neutral": 1}, "2026-04": {}, "2026-05": {"neutral": 3}, "2026-06": {"warn": 1, "recommend": 1}, "2026-07": {"neutral": 1, "recommend": 2}, "2026-08": {"compare": 2, "neutral": 3, "recommend": 3}, "2026-09": {"recommend": 5, "compare": 1, "neutral": 4, "warn": 1}, "2026-10": {"neutral": 3, "recommend": 9, "warn": 2, "compare": 2}}, "quotes": [{"id": 804546960, "item_id": 50284185, "url": "https://news.ycombinator.com/item?id=50284185", "by": "cj00", "date": "2026-10-29", "stance": "neutral", "quote": "then I use Haiku to do some categorization and auto tagging", "excerpt": "OCR on the thumbnail is sometimes good enough for me since I\u2019ll remember the teaser blurb that they put in the image but leave out of the title. So the LLM description service can handle that and maybe it\u2019ll add in \u201ctwo people standing in front of a boat\u201d and something will hit in a search. But I\u2019ve actually started pulling in more than just the thumbnail. With the YouTube API I\u2019m pulling in the description and all t", "context": "builds a semantic search project", "use_case": "categorizing and tagging video metadata", "switched_from": "", "story": "Semantic search across major museums"}, {"id": 804330768, "item_id": 50270673, "url": "https://news.ycombinator.com/item?id=50270673", "by": "wyrdcurt", "date": "2026-10-28", "stance": "recommend", "quote": "the most impressive thing about it to me was that it seemed not to overthink when it wasn't necessary", "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": 803903028, "item_id": 50243939, "url": "https://news.ycombinator.com/item?id=50243939", "by": "criley2", "date": "2026-10-26", "stance": "neutral", "quote": "I should try Haiku 5.5 high/max because it's posting higher numbers than Deepseek V4.1 flash for half the price", "excerpt": "I use Opus 5.5 at work and I use Sonnet 5.5 at home. Sonnet 5.5 medium according to AA cost per task is $0.48 per task, while GLM 5.3 (max) is $2.01 (opus 5.5 high is $1.82). GLM 5.3 flash is $0.28 per task and Deepseek 4.1 flash 1017 is $0.26. So it seems like Sonnet 5.5 medium, GLM 5.3 flash and Deepseek V4.1 flash 1017 are all in a similar space in capacity and price, however the $20 anthropic sub will get you lik", "context": "", "use_case": "", "switched_from": "", "story": "Anthropic S-1"}, {"id": 802763588, "item_id": 50172724, "url": "https://news.ycombinator.com/item?id=50172724", "by": "CharlieDigital", "date": "2026-10-20", "stance": "recommend", "quote": "Coding is handled by Haiku 5.5/GPT 6 Luna/DeepSeek v4.1 Flash.", "excerpt": "I use the frontier models only in plan mode; coding is held out to cheaper models. I run Paseo as an orchestrator with agent profiles. \"Distinguished Member of Technical Staff\" -> Fable, plan mode. \"End-User Customer Advocate\" -> Astra, plan mode. A few others. Basically, these agents (with profile prompts) can only be \"consulted\" for input on design elements; it's OK if they produce a bit of verbosity here. This has", "context": "", "use_case": "coding", "switched_from": "", "story": "SlopCodeBench 10/2026 \u2013 Opus 5.5, Astra wins overall, xhigh considered harmful"}, {"id": 802679058, "item_id": 50167441, "url": "https://news.ycombinator.com/item?id=50167441", "by": "A_D_E_P_T", "date": "2026-10-20", "stance": "recommend", "quote": "Haiku 5.5 is decent, cheap, and fast.", "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": "grunt work coding", "switched_from": "", "story": "Ask HN: Best value/money AI coding plan bundle?"}, {"id": 802555377, "item_id": 50159711, "url": "https://news.ycombinator.com/item?id=50159711", "by": "simonw", "date": "2026-10-20", "stance": "recommend", "quote": "Claude Haiku 5.5 and GPT-6 Luna - both excellent inexpensive models", "excerpt": "There are a few problems with using OpenRouter's public statistics to make statements about the wider field: - It only tracks companies that have decided to use OpenRouter to help them switch models - which biases towards companies that like open weights. Anyone using OpenAI or Anthropic directly is much less likely to be represented. - OpenRouter don't share the number of unique users represented in those numbers. I", "context": "", "use_case": "", "switched_from": "", "story": "Token-maxxing is dead, OpenRouter's token usage plateaus"}, {"id": 801866240, "item_id": 50116640, "url": "https://news.ycombinator.com/item?id=50116640", "by": "ErroneousBosh", "date": "2026-10-16", "stance": "recommend", "quote": "I've moved back to Haiku as my daily driver for a lot of stuff.", "excerpt": "I've moved back to Haiku as my daily driver for a lot of stuff. Unfortunately it doesn't really have a good video editor yet, so I'm stuck with Linux for that at least.", "context": "", "use_case": "daily driver OS", "switched_from": "", "story": "BeBox is a Haiku application that emulates Be Inc.'s BeBox"}, {"id": 800975536, "item_id": 50060971, "url": "https://news.ycombinator.com/item?id=50060971", "by": "JMKH42", "date": "2026-10-12", "stance": "recommend", "quote": "Now even Haiku at low effort does.", "excerpt": "Using the paid anthrotpic models daily, it has not been my experience that they are just fixing public examples of foolishness, but are generally getting smarter. For example I have my own private, obscure bike racing physics test I use that the newer low tier models pass now. Before fable, the top tier models didn't pass it. Now even Haiku at low effort does.", "context": "uses paid Anthropic models daily", "use_case": "", "switched_from": "", "story": "The gap in shared understanding of LLM capability is widening"}, {"id": 800204071, "item_id": 50012754, "url": "https://news.ycombinator.com/item?id=50012754", "by": "gregwebs", "date": "2026-10-08", "stance": "warn", "quote": "Haiku also doesn't seem like a competitive price/performance mix.", "excerpt": "I have been using DeepSeek 4.1 flash intensively for over a month. If I run it all day long it costs $1-2. Its fast. Previously I was always quickly running up to my Claude/Codex 5 hour window (on the $20/month plan). The cost savings of DeepSeek is real as shown in this article and I am using subsidized plans. DeepSeek is horrible at grilling sessions (the /grill* skills to make technical decisions). It doesn't know", "context": "", "use_case": "", "switched_from": "", "story": "Why isn't the industry freaking out about DeepSeek 4.1 Flash?"}, {"id": 800071282, "item_id": 50004455, "url": "https://news.ycombinator.com/item?id=50004455", "by": "zapnuk", "date": "2026-10-08", "stance": "compare", "quote": "it doesn't quite matter to me if the cheap anthropic model is 20-50% more expensive compared to luna", "excerpt": "At scale this make a difference. For me/us not that much. At my work we currently don't have a subscription and use API based billing. Some time ago the price for my usage was about 30-50\u20ac per day when we used opus for about everything. With luna+sol its about 10\u20ac for sol to plan or debug and 5\u20ac for luna to implement. At that point it doesn't quite matter to me if the cheap anthropic model is 20-50% more expensive co", "context": "", "use_case": "", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 800014818, "item_id": 50000926, "url": "https://news.ycombinator.com/item?id=50000926", "by": "zyk_computer", "date": "2026-10-08", "stance": "compare", "quote": "now I want to give it a try", "excerpt": "I used to always use glm or deepseek instead of Haiku, but now I want to give it a try.", "context": "", "use_case": "", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 799991168, "item_id": 49999448, "url": "https://news.ycombinator.com/item?id=49999448", "by": "jeremyjh", "date": "2026-10-07", "stance": "recommend", "quote": "I would mostly use Haiku in task or explorer subagents.", "excerpt": "If you can't get any coding done with 100K context that is either a broken model, a broken harness or a skill issue. I would mostly use Haiku in task or explorer subagents. I'm not saying I stay under that on every task, but I do have quite a few sessions that cap out well below that, so that price difference would be very meaningful. I use Luna for this day in and out and its excellent - if Haiku is that much better", "context": "", "use_case": "subagents", "switched_from": "Luna", "story": "Claude Haiku 5.5"}, {"id": 799967856, "item_id": 49997991, "url": "https://news.ycombinator.com/item?id=49997991", "by": "enraged_camel", "date": "2026-10-07", "stance": "recommend", "quote": "We use Haiku 4.5 to turn users' natural language queries and requests into fairly complex structured specs for interior design and construction. It has near perfect accuracy.", "excerpt": "Not really. We use Haiku 4.5 to turn users' natural language queries and requests into fairly complex structured specs for interior design and construction. It has near perfect accuracy.", "context": "builds interior design/construction product", "use_case": "natural language to structured specs", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 799958256, "item_id": 49997391, "url": "https://news.ycombinator.com/item?id=49997391", "by": "d1l", "date": "2026-10-07", "stance": "recommend", "quote": "At work we use haiku 4.5 for a handful of latency sensitive tasks that are fairly simple. It performs well.", "excerpt": "At work we use haiku 4.5 for a handful of latency sensitive tasks that are fairly simple. It performs well. Just started testing 5.5 as I\u2019ve been anticipating a nice improvement since it was teased. Results so far are trash. Prompt leakage even. And it\u2019s slower. I guess it\u2019s cheap but I think they got the balance wrong on this.", "context": "", "use_case": "latency-sensitive simple tasks", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 799958257, "item_id": 49997391, "url": "https://news.ycombinator.com/item?id=49997391", "by": "d1l", "date": "2026-10-07", "stance": "warn", "quote": "Results so far are trash. Prompt leakage even. And it\u2019s slower.", "excerpt": "At work we use haiku 4.5 for a handful of latency sensitive tasks that are fairly simple. It performs well. Just started testing 5.5 as I\u2019ve been anticipating a nice improvement since it was teased. Results so far are trash. Prompt leakage even. And it\u2019s slower. I guess it\u2019s cheap but I think they got the balance wrong on this.", "context": "", "use_case": "latency-sensitive simple tasks", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 799953684, "item_id": 49997105, "url": "https://news.ycombinator.com/item?id=49997105", "by": "wyrdcurt", "date": "2026-10-07", "stance": "warn", "quote": "If you mean Haiku 4.5 though I agree, that model was a waste of time and money.", "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": 799947376, "item_id": 49996711, "url": "https://news.ycombinator.com/item?id=49996711", "by": "mariocesar", "date": "2026-10-07", "stance": "recommend", "quote": "I use haiku for things that needs to be quick, have really clear instructions.", "excerpt": "I have a zsh functions that calls claude code with haiku to suggest commit messages, is faster and the instructions are two lines. I also have an \"ask\" script that I use daily to ask simple stuff, it can access websearch and webfetch, it's more than enough to parse logs, ask for commands, quick research on the internet, small stuff. https://github.com/mariocesar/dotfiles/blob/main/common/.local/bin/ask I use haiku fo", "context": "", "use_case": "commit messages, quick queries", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 799944640, "item_id": 49996540, "url": "https://news.ycombinator.com/item?id=49996540", "by": "giancarlostoro", "date": "2026-10-07", "stance": "neutral", "quote": "any time I ask Opus to \"try\" letting Haiku do the work, it just falls apart", "excerpt": "I with they'd give Haiku like 400k tokens roughly, I think between 400k or even 600k tokens is a sweet spot, but Haiku is basically designed to be for small edits is my understanding, but it sucks because any time I ask Opus to \"try\" letting Haiku do the work, it just falls apart and Opus comes back and tells me it switched to Sonnet (even before Sonnet finally jumped up to 5.x). I will try the new Haiku, but it woul", "context": "", "use_case": "file editing for Opus/Sonnet", "switched_from": "", "story": "Claude Haiku 5.5"}, {"id": 798340689, "item_id": 49896293, "url": "https://news.ycombinator.com/item?id=49896293", "by": "lproven", "date": "2026-09-29", "stance": "recommend", "quote": "I run Haiku, FreeBSD, OpenBSD, Ubuntu, and Pop OS, and all of them go like stink and are daily-drivable", "excerpt": "Indeed. I know of two \"best laptops ever made\": Drew DeVault says it's the X200: https://drewdevault.com/blog/Fucking-laptops/ \u00ab The best laptop ever made is the ThinkPad X200, and I have two of them. \u00bb He has a very strong case. I own one. It is certainly one of the best ever. On the other hand, Brad Taunt says it is the X220: https://btxx.org/posts/x220/ \u00ab The X220 ThinkPad is the greatest laptop ever made and you\u2019", "context": "owns ThinkPad X200/X220 machines", "use_case": "daily use on ThinkPad X220", "switched_from": "", "story": "Google ending ChromeOS support two years early"}, {"id": 798130226, "item_id": 49883139, "url": "https://news.ycombinator.com/item?id=49883139", "by": "sajithdilshan", "date": "2026-09-28", "stance": "neutral", "quote": "Haiku for running tests", "excerpt": "I use Claude Code everyday for work and the main model I use is Opus (For planning, breaking down tasks, writing tickets, implementation, etc.) and Haiku for running tests. Honestly have no idea what is the use case for Sonnet", "context": "", "use_case": "running tests", "switched_from": "", "story": "Sonnet 5.5"}, {"id": 797541088, "item_id": 49846318, "url": "https://news.ycombinator.com/item?id=49846318", "by": "qubex", "date": "2026-09-25", "stance": "recommend", "quote": "I highly recommend you give Haiku a spin", "excerpt": "I highly recommend you give Haiku a spin\u2026 the remake 25 years in the making and counting (which is several multiples the actual duration of the system it is \u201cinspired by\u201d, but that\u2019s the nature of emulators and reproductions, isn\u2019t it?). At any rate, it really gives the look and feel of the system as it was back then. The design is somehow peak late nineties distilled down into a powerful concentrate. (Make sure you ", "context": "", "use_case": "retro BeOS-like OS", "switched_from": "", "story": "Pentium II at 600Mhz with Voodoo 3 Emulated on 86Box with M6 Mac Mini"}, {"id": 797400368, "item_id": 49837523, "url": "https://news.ycombinator.com/item?id=49837523", "by": "sznio", "date": "2026-09-24", "stance": "warn", "quote": "current Haiku is useless", "excerpt": "Nothing, because current Haiku is useless. But I use Qwen3.6 35B-A3B and it's both much faster and much more capable than Haiku. And Qwen3.8 27B can do so much more, decently replaces Sonnet, albeit not as fast. Luna doesn't feel more capable nor faster than Qwen3.8. I think Anthropic could surprise us with some more innovation in this niche.", "context": "", "use_case": "", "switched_from": "", "story": "Claude Opus 5.5"}, {"id": 797244400, "item_id": 49827775, "url": "https://news.ycombinator.com/item?id=49827775", "by": "RugnirViking", "date": "2026-09-24", "stance": "neutral", "quote": "I used to use haiku for the speed. Now its just as slow as the rest.", "excerpt": "> Nobody needs to code at 600 words per second. I do. I used to use haiku for the speed. Now its just as slow as the rest. Speed is my #1 ranking of how good a model is", "context": "", "use_case": "fast coding", "switched_from": "", "story": "Mercury 2.5 LLM hits 770 tokens per second"}, {"id": 797228289, "item_id": 49826768, "url": "https://news.ycombinator.com/item?id=49826768", "by": "anshumankmr", "date": "2026-09-24", "stance": "neutral", "quote": "Absolute fucking L from anthropic for not introducing a newer cheaper Haiku4.5.", "excerpt": "Absolute fucking L from anthropic for not introducing a newer cheaper Haiku4.5. We have switched to GPT 5.6 luna (for some simpler workloads) and will switch to this too.", "context": "", "use_case": "", "switched_from": "", "story": "GPT-6 Sol and Luna"}, {"id": 796911824, "item_id": 49806989, "url": "https://news.ycombinator.com/item?id=49806989", "by": "jaapz", "date": "2026-09-22", "stance": "neutral", "quote": "Only reason I used Haiku was when I made Opus 5 run Haiku agents", "excerpt": "Only reason I used Haiku was when I made Opus 5 run Haiku agents just to rewrite Opus 5's garbage output into something legible", "context": "", "use_case": "rewriting output as subagent", "switched_from": "", "story": "Claude Opus 5.5"}, {"id": 796875744, "item_id": 49804734, "url": "https://news.ycombinator.com/item?id=49804734", "by": "enraged_camel", "date": "2026-09-22", "stance": "recommend", "quote": "It continues to be absurdly capable for converting natural language to structured JSON", "excerpt": "We use Haiku 4.5 inside our product. It continues to be absurdly capable for converting natural language to structured JSON based on a set of fairly complex business rules.", "context": "uses it inside their product", "use_case": "natural language to structured JSON", "switched_from": "", "story": "Claude Opus 5.5"}, {"id": 796072610, "item_id": 49754538, "url": "https://news.ycombinator.com/item?id=49754538", "by": "matt_oriordan", "date": "2026-09-18", "stance": "compare", "quote": "Haiku 2.5 s", "excerpt": "Hi HN, Matt here, co-founder of Ably. I built this after Jev landed on Vercel's AI Gateway: https://github.com/ably-labs/jev-pong It's pong where the ball moves one step per model decision. Four lanes, one model each, same serve, same rules. Every step the model gets the same tiny JSON state (ball, paddle, predicted intercept) and the same question: up, down or stay. One answer moves the ball one segment; eight segme", "context": "", "use_case": "Pong decision latency", "switched_from": "", "story": "Show HN: Jev vs. GPT-5.6 and Claude Haiku at Pong"}, {"id": 794776400, "item_id": 49673525, "url": "https://news.ycombinator.com/item?id=49673525", "by": "mariocesar", "date": "2026-09-12", "stance": "recommend", "quote": "To ask for web search, simple commands and research, or passing logs to analyze it's more than enough", "excerpt": "To ask for web search, simple commands and research, or passing logs to analyze it's more than enough Here is the script https://github.com/mariocesar/dotfiles/blob/main/common/.local/bin/ask", "context": "", "use_case": "web search, simple commands, log analysis", "switched_from": "", "story": "Ask HN: What default model do you use and why?"}, {"id": 794771360, "item_id": 49673210, "url": "https://news.ycombinator.com/item?id=49673210", "by": "mariocesar", "date": "2026-09-12", "stance": "recommend", "quote": "I have an 'ask' alias in my shell that just uses Haiku. I use it daily for pretty much everything", "excerpt": "I have an 'ask' alias in my shell that just uses Haiku. I use it daily for pretty much everything For more long work, I now use Fable to create a PLAN.md. I tell it to make a plan that will be executed by other models, and most of the time it ends up choosing Opus or Sonnet. I didn't start doing this recently. Before that, I would just use the top model for everything. Splitting the work across different models depen", "context": "", "use_case": "quick shell queries", "switched_from": "", "story": "Ask HN: What default model do you use and why?"}, {"id": 793502833, "item_id": 49593927, "url": "https://news.ycombinator.com/item?id=49593927", "by": "dxjxjdjsssb", "date": "2026-09-07", "stance": "recommend", "quote": "a skill/script combo suitable for Haiku to accomplish the task, and now that task happens at minimal cost", "excerpt": "I use skills for offloading work onto subagents. By configuring the skill to use a specific model it gets enforced at the harness instead of depending on the good will of the orchestrating model to actually delegate. This also saves context. Today Fable had to fetch a zip file from a web page with a eula prompt, then get at a file in a disk image in the zip. This is something that will need to happen a lot as part of", "context": "", "use_case": "cheap subagent tasks", "switched_from": "", "story": "Ask HN: How do you manage skills files?"}, {"id": 792132945, "item_id": 49508309, "url": "https://news.ycombinator.com/item?id=49508309", "by": "gen2brain", "date": "2026-08-31", "stance": "recommend", "quote": "I added Haiku because it looks and performs so well", "excerpt": "I am working on the IUP fork (UI toolkit) , with many new drivers/backends and features, and, of course, I added Haiku because it looks and performs so well, it works differently from what I used to see, making it even more interesting. The result: the absolute king in memory usage was the Win32 driver, like 4M for Hello World; in Haiku, that is 3M! For comparison, the GTK4 driver uses 40M; GTK3 and Qt6 ~15. WinUI is", "context": "", "use_case": "IUP backend target", "switched_from": "", "story": "Haiku R1/beta6 has been released"}, {"id": 792052192, "item_id": 49503262, "url": "https://news.ycombinator.com/item?id=49503262", "by": "ethin", "date": "2026-08-30", "stance": "neutral", "quote": "I would love to give Haiku a try. Indeed, I've looked at it's code before (it's quite good, or was, last time I looked at it).", "excerpt": "I would love to give Haiku a try. Indeed, I've looked at it's code before (it's quite good, or was, last time I looked at it). What prevents me is accessibility. To be fair to Haiku, implementing an actually good accessibility stack is genuinely a nightmare that AI isn't going to help much with, were you to ask it to help you with it. Primarily because many accessibility stacks (UIA, MSAA, IAccessible2, AT-SPI, ...) ", "context": "cares about accessibility, likely screen reader user", "use_case": "", "switched_from": "", "story": "Haiku R1/beta6 has been released"}, {"id": 792035840, "item_id": 49502240, "url": "https://news.ycombinator.com/item?id=49502240", "by": "jmmv", "date": "2026-08-30", "stance": "recommend", "quote": "impressed by the speed and the reasonable availability of software", "excerpt": "I tried Haiku recently on a not-super-old PC and was impressed by the speed and the reasonable availability of software. I then wished I could run this on my old Powerbook G4 because nothing else seems to run there anymore with ease, and this seems like the perfect desktop OS for such old machines. And it turns out there is now a fork for PPC: https://github.com/ActionRetro/Tabby-PPC ! (Why a fork? Because it's prima", "context": "has an old Powerbook G4", "use_case": "desktop OS", "switched_from": "", "story": "Haiku R1/beta6 has been released"}, {"id": 792024384, "item_id": 49501524, "url": "https://news.ycombinator.com/item?id=49501524", "by": "accrual", "date": "2026-08-30", "stance": "recommend", "quote": "I love the UI style of HaikuOS", "excerpt": "I love the UI style of HaikuOS, so cool to see new ports like Firefox and a Go runtime in this release. Congrats to the team!", "context": "", "use_case": "", "switched_from": "", "story": "Haiku R1/beta6 has been released"}, {"id": 792015216, "item_id": 49500951, "url": "https://news.ycombinator.com/item?id=49500951", "by": "LeFantome", "date": "2026-08-30", "stance": "compare", "quote": "But honestly Linux felt faster on this machine. The memory usage was about the same", "excerpt": "The problem is, Linux is improving too quickly for Haiku. I have been looking forward to running Haiku as a daily driver for years. And it is almost there. It may even be good enough now for many people. Modern browsers really help. But I was amazed recently when I put Haiku on an old laptop that was already running Linux. I remember Haiku being so lean and fast and multi-tasking so smoothly. But honestly Linux felt ", "context": "long-time Haiku fan", "use_case": "daily driver OS", "switched_from": "", "story": "Haiku R1/beta6 has been released"}, {"id": 790596322, "item_id": 49412270, "url": "https://news.ycombinator.com/item?id=49412270", "by": "tonyarkles", "date": "2026-08-23", "stance": "neutral", "quote": "Haiku subagents", "excerpt": "Something I\u2019ve found comparing between Fable and Opus is that Fable has impressively good analysis skills, but both of them seem to go way way overboard with \u201cpresent state\u201d comments \u201c# We\u2019re making this change here because of this issue blah blah, here\u2019s what you need to know about np.percentile, blah blah\u201d that I end up significantly pruning before making a PR. I let it do the same style verbose commit messages (be", "context": "", "use_case": "subagents", "switched_from": "", "story": "Fable and the end of the free lunch"}, {"id": 790199472, "item_id": 49387467, "url": "https://news.ycombinator.com/item?id=49387467", "by": "ricardobeat", "date": "2026-08-21", "stance": "neutral", "quote": "made a simple skill [1] that gets even Claude Haiku to write correct code", "excerpt": "See the author's more recent post \"I thought I was building a C replacement. I was wrong\": https://c3-lang.org/blog/i_thought_i_was_building_a_c_replacement/ Of note, C3 is alive and well, with the latest release (0.8.3) just last week. I stumbled upon the language earlier this year and love it. I like plain C a lot, but it's a rough experience. Having slices, modules, a strong stdlib and memory management massively ", "context": "", "use_case": "writing C3 code", "switched_from": "", "story": "The case against a C alternative (2022)"}, {"id": 786406193, "item_id": 49150387, "url": "https://news.ycombinator.com/item?id=49150387", "by": "apatheticonion", "date": "2026-08-02", "stance": "compare", "quote": "while Haiku, Opus and Sonnet are slow and create the same outputs", "excerpt": "It's because of my workflow, I think. I have my IDE open and a prompt a given task with limited scope - like \"write a function that does xyz\" \"add tests\" \"create an endpoint that accepts this json\" \"write a sql repository for this schema\" etc. I use Rust on the backend so if it compiles it works, meaning I don't need to worry about runtime errors - if it looks right, it's right. These kinds of actions don't require m", "context": "Rust backend developer", "use_case": "scoped coding tasks", "switched_from": "", "story": "Ask HN: What's Next for LLMs?"}, {"id": 785336722, "item_id": 49083545, "url": "https://news.ycombinator.com/item?id=49083545", "by": "oggreen", "date": "2026-07-28", "stance": "recommend", "quote": "at work Haiku almost always works just as well as Sonnet does and its much cheaper/faster", "excerpt": "I use Sonnet 80% of the time when I'm on claude.ai and on Claude Code my main agent is Opus and typically my sub-agents are Sonnet. I use this setup mainly because I genuinely don't know if anything I'm doing is complex enough to need Fable and I don't want to run up my usage too much. It's basically like trying to kill a gnat with a sledgehammer. Hell, at work Haiku almost always works just as well as Sonnet does an", "context": "", "use_case": "work tasks", "switched_from": "", "story": "Why I prefer Opus 5 to Fable 5"}, {"id": 782685664, "item_id": 48917854, "url": "https://news.ycombinator.com/item?id=48917854", "by": "DCKing", "date": "2026-07-15", "stance": "recommend", "quote": "I use Haiku for it, hoping that a dumber model will do a better job.", "excerpt": "I don't mind this in the agent workflows themselves, but I think this is hideous in the code and encourages agents to become even more verbose in the future. I have a workflow to create PRs where at the end, I spin up a subagent that goes through the PR to 1) remove the bulk of the comments and 2) attempt to normalize it for human consumption (I ask it to optimize wording for junior developers and non-native speakers", "context": "", "use_case": "removing slop comments from PRs", "switched_from": "", "story": "How to stop Claude from saying load-bearing"}, {"id": 774009587, "item_id": 48375599, "url": "https://news.ycombinator.com/item?id=48375599", "by": "bel8", "date": "2026-06-02", "stance": "warn", "quote": "They are cute but for serious coding they tend to waste your expensive time.", "excerpt": "It's a start and I welcome competition but I don't think I ever used small cloud models like Haiku 4.5. They are cute but for serious coding they tend to waste your expensive time. And this certainly wont bring me back to GitHub Copilot which I cancelled yesterday. GitHub Copilot had competitive pricing until yesterday when they changed from per-request to one of the most expensive per-token quotas. Seriously, take a", "context": "", "use_case": "coding", "switched_from": "", "story": "Microsoft's MAI-Code-1-Flash Scores 51% SWE-Bench Pro with Just 5B Active Params"}, {"id": 761391554, "item_id": 47586972, "url": "https://news.ycombinator.com/item?id=47586972", "by": "giancarlostoro", "date": "2026-03-31", "stance": "warn", "quote": "when I use them via JetBrains AI it feels like only Opus is good enough", "excerpt": "I'm guessing their newer models are taking way more compute than they can afford to give away. The biggest challenge of AI will eventually be, how to bring down how much compute a powerful model takes. I hope Claude puts more emphasis into making Haiku and Sonnet better, when I use them via JetBrains AI it feels like only Opus is good enough, for whatever odd reason.", "context": "", "use_case": "", "switched_from": "", "story": "Anthropic: Claude Code users hitting usage limits 'way faster than expected'"}, {"id": 750296226, "item_id": 46893514, "url": "https://news.ycombinator.com/item?id=46893514", "by": "IgorPartola", "date": "2026-02-04", "stance": "warn", "quote": "In practice the subagents took so many steps and wrote so much documentation that it became not worth it.", "excerpt": "So I have gotten pretty good at managing context such that my $20 Claude subscription rarely runs out of its quota but I still do hit it sometimes. I use Sonnet 99% of the time. Mostly this comes down to giving it specific task and using /clear frequently. I also ask it to update its own notes frequently so it doesn\u2019t have to explore the whole codebase as often. But I was really disappointed when I tried to use subag", "context": "$20 subscriber", "use_case": "subagents for routine tasks", "switched_from": "", "story": "Claude Code: connect to a local model when your quota runs out"}, {"id": 734894928, "item_id": 45930933, "url": "https://news.ycombinator.com/item?id=45930933", "by": "barefootford", "date": "2025-11-14", "stance": "recommend", "quote": "I switched from structured outputs on OpenAI apis to unstructured on Claude (haiku 4.5) and haven't had any issues (yet)", "excerpt": "I switched from structured outputs on OpenAI apis to unstructured on Claude (haiku 4.5) and haven't had any issues (yet). But guarantees are always nice.", "context": "", "use_case": "structured outputs via API", "switched_from": "OpenAI API", "story": "Structured outputs on the Claude Developer Platform"}]}