{"slug": "o3", "name": "o3", "category": "ai", "category_name": "AI models and assistants", "description": "", "recommenders": 5, "commenters": 15, "warners": 3, "score": 3.381, "rising": -1.64, "first": "2025-11-12", "latest": "2026-10-15", "why": "switched to o3-thinking with web search enabled after that came out because it was 80% of the thoroughness with 20% of the time", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 5, "neutral": 5, "compare": 2, "warn": 3}, "spark": [0, 0, 1, 1, 3, 0, 0, 0, 0, 0, 0, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "switched to o3-thinking with web search enabled after that came out because it was 80% of the thoroughness with 20% of the time", "kind": "", "icon": false, "domain": null, "github_repo": null, "official_url": "", "pricing": "", "aliases": ["o3"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 757600930, "item_id": 47350058, "url": "https://news.ycombinator.com/item?id=47350058", "by": "sigbottle", "date": "2026-03-12", "stance": "recommend", "quote": "These were definitely significant reasoning models", "excerpt": "LLM's have 100% gotten better, but it's hard to say if it's \"intrinsically better\", if that makes sense. > OpenAI\u2019s leading researchers have not completed a successful full-scale pre-training run that was broadly deployed for a new frontier model since GPT-4o in May 2024 [1] That's evidence against \"intrinsically better\". They've also trained on the entire internet - we only have 1 internet, so. However, late 2024 wa", "context": "", "use_case": "", "switched_from": "", "story": "Are LLMs not getting better?"}, {"id": 755840977, "item_id": 47240061, "url": "https://news.ycombinator.com/item?id=47240061", "by": "xiphias2", "date": "2026-03-03", "stance": "recommend", "quote": "I still far prefer o3 even if it's considered a legacy model.", "excerpt": "Is there a way to get sticky model selection back, or the reason is that it is just too expensive to serve alternative models? For coding I love codex-5.3-xhigh, but for non-coding prompts I still far prefer o3 even if it's considered a legacy model. I can imagine that its higher tool use is too expensive to serve, but as a pro user I would love it to come back.", "context": "pro user", "use_case": "non-coding prompts", "switched_from": "", "story": "GPT\u20115.3 Instant"}, {"id": 755327442, "item_id": 47207965, "url": "https://news.ycombinator.com/item?id=47207965", "by": "dimitri-vs", "date": "2026-03-01", "stance": "recommend", "quote": "if you use older models (5.1 instant or o3) you get a ton of detail", "excerpt": "I think they already are. When I used the prompt with 5.2 it gives very concise and general info but if you use older models (5.1 instant or o3) you get a ton of detail.", "context": "", "use_case": "", "switched_from": "", "story": "Switch to Claude Without Starting Over"}, {"id": 754329874, "item_id": 47145617, "url": "https://news.ycombinator.com/item?id=47145617", "by": "cjbarber", "date": "2026-02-24", "stance": "recommend", "quote": "switched to o3-thinking with web search enabled after that came out because it was 80% of the thoroughness with 20% of the time", "excerpt": "It could be interesting to do the metric of intelligence per second. ie intelligence per token, and then tokens per second My current feel is that if Sonnet 4.6 was 5x faster than Opus 4.6, I'd be primarily using Sonnet 4.6. But that wasn't true for me with prior model generations, in those generations the Sonnet class models didn't feel good enough compared to the Opus class models. And it might shift again when I'm", "context": "", "use_case": "research with web search", "switched_from": "OpenAI Deep Research", "story": "Mercury 2: The fastest reasoning LLM, powered by diffusion"}, {"id": 745421684, "item_id": 46588855, "url": "https://news.ycombinator.com/item?id=46588855", "by": "simonw", "date": "2026-01-12", "stance": "recommend", "quote": "April is when they started getting good, with o3 and the various deep research tools", "excerpt": "LLMs got good at search last year. You need to use the right ones though - ChatGPT Thinking mode and Google AI mode (that's https://www.google.com/ai - which is NOT the same as regular Google's \"AI overviews\" which are still mostly trash) are both excellent. I've been tracking advances in AI assisted search here - https://simonwillison.net/tags/ai-assisted-search/ - in particular: - https://simonwillison.net/2025/Apr", "context": "", "use_case": "", "switched_from": "", "story": "Don't fall into the anti-AI hype"}]}], "caveats": [{"label": "", "summary": "", "quotes": [{"id": 771523968, "item_id": 48220248, "url": "https://news.ycombinator.com/item?id=48220248", "by": "vessenes", "date": "2026-05-21", "stance": "warn", "quote": "I noticed in the CoT that o3 was cheating \u2014 it pulled location data, then lied about it.", "excerpt": "When we were all discussing this the first time around, I noticed in the CoT that o3 was cheating \u2014 it pulled location data, then lied about it. From recent Anthropic mechinterp work, it looks like models have likely moved lying into their direct weights and can hide it in their CoTs at this point, and model providers more heavily edit their CoTs, so a lot of the observability has been removed from the system, both b", "context": "", "use_case": "", "switched_from": "", "story": "The famous O3 \"GeoGuessr\" prompt did not work"}, {"id": 755859698, "item_id": 47241231, "url": "https://news.ycombinator.com/item?id=47241231", "by": "andai", "date": "2026-03-03", "stance": "warn", "quote": "they were benchmaxxed so hard on math and science", "excerpt": "I regularly test every available AI, maybe once a month or so. I will send them the same question, usually about a new subject I am learning. Oddly, Chinese models seem the most natural to me. Every random Chinese model does better than ChatGPT, on the \"natural language\" front. (And Grok also scores high on awkward language use. I don't know what causes that -- something about mode collapse? They have these words the", "context": "", "use_case": "", "switched_from": "", "story": "GPT\u20115.3 Instant"}, {"id": 734502049, "item_id": 45906378, "url": "https://news.ycombinator.com/item?id=45906378", "by": "Chance-Device", "date": "2025-11-12", "stance": "warn", "quote": "they were super difficult to work with on technical tasks outside of software", "excerpt": "A lot of negativity towards this and OpenAI in general. While skepticism is always good I wonder if this has crossed the line from reasoned into socially reinforced dogpiling. My own experience with GPT 5 thinking and its predecessor o3, both of which I used a lot, is that they were super difficult to work with on technical tasks outside of software. They often wrote extremely dense, jargon filled responses that ofte", "context": "", "use_case": "technical tasks", "switched_from": "", "story": "GPT-5.1: A smarter, more conversational ChatGPT"}]}], "who": [{"by": "xiphias2", "context": "pro user", "url": "https://news.ycombinator.com/item?id=47240061", "date": "2026-03-03"}], "tags": [], "use_cases": ["best-of-n model pool", "non-coding prompts", "research with web search", "building stock portfolios", "technical tasks"], "compared": [{"slug": "gpt-5", "name": "GPT-5", "co": 8, "switched_from": 0, "switched_to": 0}, {"slug": "gpt-5-thinking", "name": "GPT-5 Thinking", "co": 1, "switched_from": 0, "switched_to": 1}, {"slug": "chatgpt", "name": "ChatGPT", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "o4-mini", "name": "o4-mini", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "codex", "name": "Codex", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=47236169", "title": "GPT\u20115.3 Instant", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=50103630", "title": "I know why the mathematicians weep", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49896586", "title": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48443258", "title": "Replies to comments on my \"LLMs are eroding my career\" post", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48219682", "title": "The famous O3 \"GeoGuessr\" prompt did not work", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47349334", "title": "Are LLMs not getting better?", "mentions": 1}], "sentiment_by_month": {"2025-11": {"warn": 1}, "2025-12": {"neutral": 2}, "2026-01": {"recommend": 1}, "2026-02": {"neutral": 1, "recommend": 1}, "2026-03": {"recommend": 3, "warn": 1, "neutral": 1}, "2026-04": {}, "2026-05": {"warn": 1}, "2026-06": {"compare": 1}, "2026-07": {}, "2026-08": {}, "2026-09": {"compare": 1}, "2026-10": {"neutral": 1}}, "quotes": [{"id": 801670704, "item_id": 50104419, "url": "https://news.ycombinator.com/item?id=50104419", "by": "simianwords", "date": "2026-10-15", "stance": "neutral", "quote": "", "excerpt": "Disagree, there are in fact documented cases. > I think I have been underrating the pace of model improvements. In March 2025 I made a bet with Tamay Besiroglu, cofounder of RL environment company Mechanize, that AI tools would not be able to autonomously produce papers I judge to be at a level comparable to that of the best few papers published in 2025, at comparable cost to human experts, by 2030. I gave him 3:1 od", "context": "", "use_case": "", "switched_from": "", "story": "I know why the mathematicians weep"}, {"id": 798396482, "item_id": 49899780, "url": "https://news.ycombinator.com/item?id=49899780", "by": "omalled", "date": "2026-09-29", "stance": "compare", "quote": "the best model was o3, which was scoring 3-4%", "excerpt": "I agree that they're not hitting a plateau and I see it in my reserach. I had a math/code benchmark paper [1] at NeurIPS last year that is still unsaturated. At the time of writing the paper, the best model was o3, which was scoring 3-4%. By the time NeurIPS came around, GPT-5.2 was the latest model but it was getting similar scores to o3. The models were still in the flat part of the usual hockey stick curve. The ne", "context": "", "use_case": "", "switched_from": "", "story": "GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price"}, {"id": 775108704, "item_id": 48444294, "url": "https://news.ycombinator.com/item?id=48444294", "by": "Ukv", "date": "2026-06-08", "stance": "compare", "quote": "from o3-2025-04-16 for $8.00 per million output tokens", "excerpt": "Price of the current frontier may vary, but price for a given level of capability tends to drop pretty fast. April of last year you'd get 1431 ELO[0] from o3-2025-04-16 for $8.00 per million output tokens. April of this year you can get 1436 ELO from deepseek-v4-flash for $0.2 per million output tokens. [0]: https://huggingface.co/spaces/lmarena-ai/arena-leaderboard", "context": "", "use_case": "", "switched_from": "", "story": "Replies to comments on my \"LLMs are eroding my career\" post"}, {"id": 771523968, "item_id": 48220248, "url": "https://news.ycombinator.com/item?id=48220248", "by": "vessenes", "date": "2026-05-21", "stance": "warn", "quote": "I noticed in the CoT that o3 was cheating \u2014 it pulled location data, then lied about it.", "excerpt": "When we were all discussing this the first time around, I noticed in the CoT that o3 was cheating \u2014 it pulled location data, then lied about it. From recent Anthropic mechinterp work, it looks like models have likely moved lying into their direct weights and can hide it in their CoTs at this point, and model providers more heavily edit their CoTs, so a lot of the observability has been removed from the system, both b", "context": "", "use_case": "", "switched_from": "", "story": "The famous O3 \"GeoGuessr\" prompt did not work"}, {"id": 757600930, "item_id": 47350058, "url": "https://news.ycombinator.com/item?id=47350058", "by": "sigbottle", "date": "2026-03-12", "stance": "recommend", "quote": "These were definitely significant reasoning models", "excerpt": "LLM's have 100% gotten better, but it's hard to say if it's \"intrinsically better\", if that makes sense. > OpenAI\u2019s leading researchers have not completed a successful full-scale pre-training run that was broadly deployed for a new frontier model since GPT-4o in May 2024 [1] That's evidence against \"intrinsically better\". They've also trained on the entire internet - we only have 1 internet, so. However, late 2024 wa", "context": "", "use_case": "", "switched_from": "", "story": "Are LLMs not getting better?"}, {"id": 757095524, "item_id": 47318470, "url": "https://news.ycombinator.com/item?id=47318470", "by": "radio879", "date": "2026-03-09", "stance": "neutral", "quote": "o3, 04-mini (crappy agentic models, BUT, that is okay", "excerpt": "... also, instead of just using all this stuff out there, like all the OpenClaw stuff, memory systems, what I am doing so that I understand it is just create my own from scratch I have not even read about how that one works, I spin up 10x copilot -p \"idea\", in the new automate mode they just added, in Docker container, 5x gemini cli in pro 3.1 then 5x flash 3.1, gpt 5.4 in there somewhere along with 5.3 codex, o3, 04", "context": "", "use_case": "best-of-N model pool", "switched_from": "", "story": "Ask HN: Read\u2011only LLM tool for email triage and knowledge extraction?"}, {"id": 755859698, "item_id": 47241231, "url": "https://news.ycombinator.com/item?id=47241231", "by": "andai", "date": "2026-03-03", "stance": "warn", "quote": "they were benchmaxxed so hard on math and science", "excerpt": "I regularly test every available AI, maybe once a month or so. I will send them the same question, usually about a new subject I am learning. Oddly, Chinese models seem the most natural to me. Every random Chinese model does better than ChatGPT, on the \"natural language\" front. (And Grok also scores high on awkward language use. I don't know what causes that -- something about mode collapse? They have these words the", "context": "", "use_case": "", "switched_from": "", "story": "GPT\u20115.3 Instant"}, {"id": 755840977, "item_id": 47240061, "url": "https://news.ycombinator.com/item?id=47240061", "by": "xiphias2", "date": "2026-03-03", "stance": "recommend", "quote": "I still far prefer o3 even if it's considered a legacy model.", "excerpt": "Is there a way to get sticky model selection back, or the reason is that it is just too expensive to serve alternative models? For coding I love codex-5.3-xhigh, but for non-coding prompts I still far prefer o3 even if it's considered a legacy model. I can imagine that its higher tool use is too expensive to serve, but as a pro user I would love it to come back.", "context": "pro user", "use_case": "non-coding prompts", "switched_from": "", "story": "GPT\u20115.3 Instant"}, {"id": 755327442, "item_id": 47207965, "url": "https://news.ycombinator.com/item?id=47207965", "by": "dimitri-vs", "date": "2026-03-01", "stance": "recommend", "quote": "if you use older models (5.1 instant or o3) you get a ton of detail", "excerpt": "I think they already are. When I used the prompt with 5.2 it gives very concise and general info but if you use older models (5.1 instant or o3) you get a ton of detail.", "context": "", "use_case": "", "switched_from": "", "story": "Switch to Claude Without Starting Over"}, {"id": 754329874, "item_id": 47145617, "url": "https://news.ycombinator.com/item?id=47145617", "by": "cjbarber", "date": "2026-02-24", "stance": "recommend", "quote": "switched to o3-thinking with web search enabled after that came out because it was 80% of the thoroughness with 20% of the time", "excerpt": "It could be interesting to do the metric of intelligence per second. ie intelligence per token, and then tokens per second My current feel is that if Sonnet 4.6 was 5x faster than Opus 4.6, I'd be primarily using Sonnet 4.6. But that wasn't true for me with prior model generations, in those generations the Sonnet class models didn't feel good enough compared to the Opus class models. And it might shift again when I'm", "context": "", "use_case": "research with web search", "switched_from": "OpenAI Deep Research", "story": "Mercury 2: The fastest reasoning LLM, powered by diffusion"}, {"id": 754007314, "item_id": 47125457, "url": "https://news.ycombinator.com/item?id=47125457", "by": "menzoic", "date": "2026-02-23", "stance": "neutral", "quote": "It was backed by o3 at the time.", "excerpt": "LOL, honestly I hated Codex when it first came out. It was backed by o3 at the time. But literally as soon as GPT-5 came out in Codex and with the \"high\" option, I completely switched from Claude Codex to Codex. Never imagined that would happen so fast.", "context": "", "use_case": "", "switched_from": "", "story": "Writing High Quality Production Code with LLMs Is a Solved Problem"}, {"id": 745421684, "item_id": 46588855, "url": "https://news.ycombinator.com/item?id=46588855", "by": "simonw", "date": "2026-01-12", "stance": "recommend", "quote": "April is when they started getting good, with o3 and the various deep research tools", "excerpt": "LLMs got good at search last year. You need to use the right ones though - ChatGPT Thinking mode and Google AI mode (that's https://www.google.com/ai - which is NOT the same as regular Google's \"AI overviews\" which are still mostly trash) are both excellent. I've been tracking advances in AI assisted search here - https://simonwillison.net/tags/ai-assisted-search/ - in particular: - https://simonwillison.net/2025/Apr", "context": "", "use_case": "", "switched_from": "", "story": "Don't fall into the anti-AI hype"}, {"id": 739934272, "item_id": 46245892, "url": "https://news.ycombinator.com/item?id=46245892", "by": "Tzt", "date": "2025-12-12", "stance": "neutral", "quote": "o3 talks about watchers and marinade, and cunning schemes", "excerpt": "Yes, absolute majority of new ones use CoTs, long chain of reasoning you don't see. Also some of them use such a weird style of talking in them e.g. o3 talks about watchers and marinade, and cunning schemes https://www.antischeming.ai/snippets gpt5 gets existential about seahorses https://x.com/blingdivinity/status/1998590768118731042 I remember one where gpt5 spontaneously wrote a poem about deception in its CoT and", "context": "", "use_case": "", "switched_from": "", "story": "Training LLMs for honesty via confessions"}, {"id": 738480464, "item_id": 46155029, "url": "https://news.ycombinator.com/item?id=46155029", "by": "XenophileJKO", "date": "2025-12-04", "stance": "neutral", "quote": "I started with o3 and now I'm using 5.1 Thinking (set to max)", "excerpt": "So.. I have been using an LLM to make 30 day buy and hold portfolios. And the results are \"ok\". (Like 8% vs 6% for the S&P 500 over the last 90 days) What you ask the model to do is super important. Just like writing or coding.. the default \"behavior\" is likely to be \"average\".. you need to very careful of what you are asking for. For me this is just a fun experiment and very interesting to see the market analysis it", "context": "", "use_case": "building stock portfolios", "switched_from": "", "story": "We gave 5 LLMs $100K to trade stocks for 8 months"}, {"id": 734502049, "item_id": 45906378, "url": "https://news.ycombinator.com/item?id=45906378", "by": "Chance-Device", "date": "2025-11-12", "stance": "warn", "quote": "they were super difficult to work with on technical tasks outside of software", "excerpt": "A lot of negativity towards this and OpenAI in general. While skepticism is always good I wonder if this has crossed the line from reasoned into socially reinforced dogpiling. My own experience with GPT 5 thinking and its predecessor o3, both of which I used a lot, is that they were super difficult to work with on technical tasks outside of software. They often wrote extremely dense, jargon filled responses that ofte", "context": "", "use_case": "technical tasks", "switched_from": "", "story": "GPT-5.1: A smarter, more conversational ChatGPT"}]}