{"slug": "gpt-5-codex", "name": "GPT-5 Codex", "category": "ai-coding", "category_name": "AI coding tools", "description": "OpenAI's line of GPT models tuned for agentic software coding tasks.", "recommenders": 7, "commenters": 18, "warners": 6, "score": 2.163, "rising": -2.29, "first": "2025-11-08", "latest": "2026-08-07", "why": "Commenters use it as a cost-effective daily coding workhorse and for thorough, narrowly scoped work like debugging and code review.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 7, "neutral": 5, "compare": 0, "warn": 6}, "spark": [1, 0, 1, 0, 1, 0, 1, 3, 0, 0, 0, 0], "listed": true, "loved": ["Value for cost", "Daily driver"], "caveat": "Quality regressions", "short_quote": "5.3-codex might be the best overall coding model OpenAI have ever released. It's 90% as good as 5.5 and costs about 20% as much", "icon": true, "domain": null, "github_repo": null, "official_url": "https://openai.com/codex/", "pricing": "paid", "aliases": ["GPT-5.3 Codex", "GPT-5.3-Codex", "GPT-5.5 Codex", "GPT 5.3 Codex", "GPT-5.2-Codex", "GPT-5 Codex", "GPT-5.1 Codex", "GPT 5.3-codex", "GPT 5.2 Codex", "GPT 5.1 Codex", "GPT-5.2 Codex"], "verdict": [{"text": "Commenters use it as a cost-effective daily coding workhorse", "cites": [{"id": 779824402, "item_id": 48739025, "url": "https://news.ycombinator.com/item?id=48739025", "by": "ifwinterco", "date": "2026-06-30", "stance": "recommend", "quote": "5.3-codex might be the best overall coding model OpenAI have ever released. It's 90% as good as 5.5 and costs about 20% as much", "excerpt": "I prefer GPT 5.5 to Opus but both are absurdly expensive token hogs, I can't afford to use either as my main model at $work with the monthly spend cap we have. I use Composer (since we use Cursor) or GPT 5.3-codex as my workhorse models and only break out the big guns when I have a genuinely difficult problem to solve. IMO somewhat weirdly 5.3-codex might be the best overall coding model OpenAI have ever released. It", "context": "works under a monthly spend cap at work", "use_case": "main coding workhorse model", "switched_from": "", "story": "Claude Sonnet 5"}, {"id": 770924432, "item_id": 48182777, "url": "https://news.ycombinator.com/item?id=48182777", "by": "dgunay", "date": "2026-05-18", "stance": "recommend", "quote": "For straightforward coding tasks I use gpt-5.3-codex on high or xhigh.", "excerpt": "For straightforward coding tasks I use gpt-5.3-codex on high or xhigh. Sometimes I try 5.5 but overall 5.3-codex is more than capable enough for most of my needs and quite a bit cheaper. For more interactive/discussion/planning or orchestration stuff, I find myself going back and forth between Opus 4.7 and GPT 5.5. Still not sure which one I prefer.", "context": "", "use_case": "straightforward coding tasks", "switched_from": "", "story": "Ask HN: What LLM models are you using and why?"}, {"id": 777731424, "item_id": 48608214, "url": "https://news.ycombinator.com/item?id=48608214", "by": "wiether", "date": "2026-06-20", "stance": "recommend", "quote": "Since Codex-5.3 came out it was my daily driver for everything", "excerpt": "Since Codex-5.3 came out it was my daily driver for everything: quick scripting, greenfield projects, new features on old projects... Idk if it was the harness (OpenCode), my AGENT or my prompts, but I was getting exactly what I wanted, and quickly. With GPT-5.5 it tries to play smart, takes much more times and is often stuck on basic stuff that DeepSeek solves oneshot.", "context": "", "use_case": "scripting, greenfield and feature work", "switched_from": "", "story": "GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2"}]}, {"text": " and for thorough, narrowly scoped work like debugging and code review.", "cites": [{"id": 736670641, "item_id": 46041915, "url": "https://news.ycombinator.com/item?id=46041915", "by": "nickandbro", "date": "2025-11-24", "stance": "recommend", "quote": "5.1 Codex I use for a narrowly defined task where I can just fire and forget it.", "excerpt": "I use the following models like so nowadays: Gemini is great, when you have gitingested the code of pypi package and want to use it as context. This comes in handy for tasks and repos outside the model's training data. 5.1 Codex I use for a narrowly defined task where I can just fire and forget it. For example, codex will troubleshoot why a websocket is not working, by running its own curl requests within cursor or e", "context": "", "use_case": "fire-and-forget debugging tasks", "switched_from": "", "story": "Claude Opus 4.5"}, {"id": 777144464, "item_id": 48571529, "url": "https://news.ycombinator.com/item?id=48571529", "by": "vunderba", "date": "2026-06-17", "stance": "recommend", "quote": "gpt-5.5 codex xhigh reasoning seemed to catch more edge cases and went \"deeper\" into analysis than Opus 4.7/4.8.", "excerpt": "I use LLMs more in the context of peer-reviewing and also came to a similar conclusion, gpt-5.5 codex xhigh reasoning seemed to catch more edge cases and went \"deeper\" into analysis than Opus 4.7/4.8. My preliminary tests of Fable were pretty promising but that's DOA for everyone for now.", "context": "", "use_case": "peer-reviewing", "switched_from": "", "story": "Leaked OpenAI financials show $38.5B loss and compute burn"}]}, {"text": " The most common complaint is inconsistent quality,", "cites": [{"id": 780637536, "item_id": 48789846, "url": "https://news.ycombinator.com/item?id=48789846", "by": "zenapollo", "date": "2026-07-04", "stance": "warn", "quote": "I\u2019m seeing incredibly stupid implementations intermittently", "excerpt": "I\u2019ve definitely experienced step jumps down in quality on an almost daily basis. I usually used xhigh. The experience of relying on codex\u2019s outstandingly thorough coding earlier in the year has evaporated for me. I\u2019m seeing incredibly stupid implementations intermittently, and have simply switched to Claude until openai takes the issue seriously. As far as i could tell they haven\u2019t taken it seriously for the several ", "context": "", "use_case": "coding", "switched_from": "", "story": "GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance"}, {"id": 751751521, "item_id": 46984470, "url": "https://news.ycombinator.com/item?id=46984470", "by": "muzani", "date": "2026-02-11", "stance": "warn", "quote": "gpt-5.3-codex benchmarked better, but I'm not seeing this translating to useful code. It failed with the first few requests I gave it.", "excerpt": "By \"here\", certainly you don't mean HN? HN feels muted and honestly, I haven't been on here because it's quite stale by the time it gets here, and people have just been vocal about not getting it. > AI.com was sold for $70M This news is everywhere elsewhere, and it's still going on in my WhatsApp groups. I can't even find the active HN thread. This was more because people were surprised about the actual owner of ai.c", "context": "", "use_case": "coding", "switched_from": "", "story": "Ask HN: Why is everyone here so AI-hyped?"}]}, {"text": " with reports of intermittently poor or hallucinated output.", "cites": [{"id": 780637536, "item_id": 48789846, "url": "https://news.ycombinator.com/item?id=48789846", "by": "zenapollo", "date": "2026-07-04", "stance": "warn", "quote": "I\u2019m seeing incredibly stupid implementations intermittently", "excerpt": "I\u2019ve definitely experienced step jumps down in quality on an almost daily basis. I usually used xhigh. The experience of relying on codex\u2019s outstandingly thorough coding earlier in the year has evaporated for me. I\u2019m seeing incredibly stupid implementations intermittently, and have simply switched to Claude until openai takes the issue seriously. As far as i could tell they haven\u2019t taken it seriously for the several ", "context": "", "use_case": "coding", "switched_from": "", "story": "GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance"}, {"id": 785947720, "item_id": 49121732, "url": "https://news.ycombinator.com/item?id=49121732", "by": "user43928", "date": "2026-07-31", "stance": "warn", "quote": "\"I love hallucinated garbage\": Sonnet, Qwen 3.6, GPT 5.4 mini, GPT 5.3 Codex", "excerpt": "In my mind there are three tiers: The SOTA: Fable, GPT 5.6 Sol, Opus 5 The \"enterprise admin did not turn on the new models\": Opus 4.8, GPT 5.5 The \"I love hallucinated garbage\": Sonnet, Qwen 3.6, GPT 5.4 mini, GPT 5.3 Codex, etc. Results vary widely", "context": "", "use_case": "", "switched_from": "", "story": "Google fixed more Chrome bugs in June than over the past two years, thanks to AI"}]}], "praise": [{"label": "Value for cost", "summary": "Described as nearly as capable as pricier models while costing far less, in part because it uses fewer tokens.", "quotes": [{"id": 779824402, "item_id": 48739025, "url": "https://news.ycombinator.com/item?id=48739025", "by": "ifwinterco", "date": "2026-06-30", "stance": "recommend", "quote": "5.3-codex might be the best overall coding model OpenAI have ever released. It's 90% as good as 5.5 and costs about 20% as much", "excerpt": "I prefer GPT 5.5 to Opus but both are absurdly expensive token hogs, I can't afford to use either as my main model at $work with the monthly spend cap we have. I use Composer (since we use Cursor) or GPT 5.3-codex as my workhorse models and only break out the big guns when I have a genuinely difficult problem to solve. IMO somewhat weirdly 5.3-codex might be the best overall coding model OpenAI have ever released. It", "context": "works under a monthly spend cap at work", "use_case": "main coding workhorse model", "switched_from": "", "story": "Claude Sonnet 5"}, {"id": 770924432, "item_id": 48182777, "url": "https://news.ycombinator.com/item?id=48182777", "by": "dgunay", "date": "2026-05-18", "stance": "recommend", "quote": "For straightforward coding tasks I use gpt-5.3-codex on high or xhigh.", "excerpt": "For straightforward coding tasks I use gpt-5.3-codex on high or xhigh. Sometimes I try 5.5 but overall 5.3-codex is more than capable enough for most of my needs and quite a bit cheaper. For more interactive/discussion/planning or orchestration stuff, I find myself going back and forth between Opus 4.7 and GPT 5.5. Still not sure which one I prefer.", "context": "", "use_case": "straightforward coding tasks", "switched_from": "", "story": "Ask HN: What LLM models are you using and why?"}]}, {"label": "Daily driver", "summary": "Used as the main model for scripting, greenfield projects and routine coding tasks.", "quotes": [{"id": 777731424, "item_id": 48608214, "url": "https://news.ycombinator.com/item?id=48608214", "by": "wiether", "date": "2026-06-20", "stance": "recommend", "quote": "Since Codex-5.3 came out it was my daily driver for everything", "excerpt": "Since Codex-5.3 came out it was my daily driver for everything: quick scripting, greenfield projects, new features on old projects... Idk if it was the harness (OpenCode), my AGENT or my prompts, but I was getting exactly what I wanted, and quickly. With GPT-5.5 it tries to play smart, takes much more times and is often stuck on basic stuff that DeepSeek solves oneshot.", "context": "", "use_case": "scripting, greenfield and feature work", "switched_from": "", "story": "GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2"}, {"id": 770924432, "item_id": 48182777, "url": "https://news.ycombinator.com/item?id=48182777", "by": "dgunay", "date": "2026-05-18", "stance": "recommend", "quote": "For straightforward coding tasks I use gpt-5.3-codex on high or xhigh.", "excerpt": "For straightforward coding tasks I use gpt-5.3-codex on high or xhigh. Sometimes I try 5.5 but overall 5.3-codex is more than capable enough for most of my needs and quite a bit cheaper. For more interactive/discussion/planning or orchestration stuff, I find myself going back and forth between Opus 4.7 and GPT 5.5. Still not sure which one I prefer.", "context": "", "use_case": "straightforward coding tasks", "switched_from": "", "story": "Ask HN: What LLM models are you using and why?"}, {"id": 779824402, "item_id": 48739025, "url": "https://news.ycombinator.com/item?id=48739025", "by": "ifwinterco", "date": "2026-06-30", "stance": "recommend", "quote": "5.3-codex might be the best overall coding model OpenAI have ever released. It's 90% as good as 5.5 and costs about 20% as much", "excerpt": "I prefer GPT 5.5 to Opus but both are absurdly expensive token hogs, I can't afford to use either as my main model at $work with the monthly spend cap we have. I use Composer (since we use Cursor) or GPT 5.3-codex as my workhorse models and only break out the big guns when I have a genuinely difficult problem to solve. IMO somewhat weirdly 5.3-codex might be the best overall coding model OpenAI have ever released. It", "context": "works under a monthly spend cap at work", "use_case": "main coding workhorse model", "switched_from": "", "story": "Claude Sonnet 5"}]}, {"label": "Thorough analysis", "summary": "Credited with catching more edge cases and going deeper in reviews, and with thorough coding.", "quotes": [{"id": 777144464, "item_id": 48571529, "url": "https://news.ycombinator.com/item?id=48571529", "by": "vunderba", "date": "2026-06-17", "stance": "recommend", "quote": "gpt-5.5 codex xhigh reasoning seemed to catch more edge cases and went \"deeper\" into analysis than Opus 4.7/4.8.", "excerpt": "I use LLMs more in the context of peer-reviewing and also came to a similar conclusion, gpt-5.5 codex xhigh reasoning seemed to catch more edge cases and went \"deeper\" into analysis than Opus 4.7/4.8. My preliminary tests of Fable were pretty promising but that's DOA for everyone for now.", "context": "", "use_case": "peer-reviewing", "switched_from": "", "story": "Leaked OpenAI financials show $38.5B loss and compute burn"}, {"id": 780637536, "item_id": 48789846, "url": "https://news.ycombinator.com/item?id=48789846", "by": "zenapollo", "date": "2026-07-04", "stance": "warn", "quote": "I\u2019m seeing incredibly stupid implementations intermittently", "excerpt": "I\u2019ve definitely experienced step jumps down in quality on an almost daily basis. I usually used xhigh. The experience of relying on codex\u2019s outstandingly thorough coding earlier in the year has evaporated for me. I\u2019m seeing incredibly stupid implementations intermittently, and have simply switched to Claude until openai takes the issue seriously. As far as i could tell they haven\u2019t taken it seriously for the several ", "context": "", "use_case": "coding", "switched_from": "", "story": "GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance"}]}, {"label": "Autonomous debugging", "summary": "Trusted with narrowly defined tasks it can run itself, such as debugging inside containers.", "quotes": [{"id": 736670641, "item_id": 46041915, "url": "https://news.ycombinator.com/item?id=46041915", "by": "nickandbro", "date": "2025-11-24", "stance": "recommend", "quote": "5.1 Codex I use for a narrowly defined task where I can just fire and forget it.", "excerpt": "I use the following models like so nowadays: Gemini is great, when you have gitingested the code of pypi package and want to use it as context. This comes in handy for tasks and repos outside the model's training data. 5.1 Codex I use for a narrowly defined task where I can just fire and forget it. For example, codex will troubleshoot why a websocket is not working, by running its own curl requests within cursor or e", "context": "", "use_case": "fire-and-forget debugging tasks", "switched_from": "", "story": "Claude Opus 4.5"}, {"id": 753817987, "item_id": 47113624, "url": "https://news.ycombinator.com/item?id=47113624", "by": "stared", "date": "2026-02-22", "stance": "recommend", "quote": "GPT-5.2-Codex worked with it well!", "excerpt": "Thanks for sharing! It seems to be an active space, vide a recent MCP server (https://news.ycombinator.com/item?id=46882389). I you haven't tried, recommend a lot posting it as Show HN. I tried a few approaches - https://github.com/jtang613/GhidrAssistMCP (was the harderst to set) Ghidra analyzeHeadless (GPT-5.2-Codex worked with it well!) and PyGhidra (my go-to). Did you try to see which works the best? I mean, very", "context": "", "use_case": "driving Ghidra analysis", "switched_from": "", "story": "We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them"}]}, {"label": "Large codebases", "summary": "Reported to handle large, complex codebases better than some alternatives.", "quotes": [{"id": 747416851, "item_id": 46713553, "url": "https://news.ycombinator.com/item?id=46713553", "by": "Octokat", "date": "2026-01-21", "stance": "recommend", "quote": "the other 2 do a much better job at the level and complexity of large codebases", "excerpt": "I use Cursor with a max subscription through work and using Gemini in multi-model mode alongside Opus 4.5 and GPT 5.2 Codex, Gemini's been the worst performer. Not saying it's a bad model just that the other 2 do a much better job at the level and complexity of large codebases. Just my experience.", "context": "", "use_case": "large codebase coding", "switched_from": "", "story": "Ask HN: What single AI tool/technique 10x'd your productivity last year?"}, {"id": 756410288, "item_id": 47275643, "url": "https://news.ycombinator.com/item?id=47275643", "by": "polyglotfacto", "date": "2026-03-06", "stance": "recommend", "quote": "Also switched to a more expensive coding model (GPT 5.3 Codex), which was an improvement", "excerpt": "I am in the third week of my experiment with AI-assisted coding on a personal branch of Servo (which still bans all AI contributions). First, the least relevant metric: LOC. Almost doubled from 7k to 12k in a week. Current WPT pass rate, compared to the default integration of rustnn (pywebnn) is at: https://gist.github.com/gterzian/95371d56468390177982c068480db95d This week was the grunt work: grinding through the lo", "context": "experimenting with AI-assisted coding on a Servo branch", "use_case": "AI-assisted coding", "switched_from": "", "story": null}]}], "caveats": [{"label": "Quality regressions", "summary": "Some users report quality dropping over time, with intermittently poor implementations.", "quotes": [{"id": 780637536, "item_id": 48789846, "url": "https://news.ycombinator.com/item?id=48789846", "by": "zenapollo", "date": "2026-07-04", "stance": "warn", "quote": "I\u2019m seeing incredibly stupid implementations intermittently", "excerpt": "I\u2019ve definitely experienced step jumps down in quality on an almost daily basis. I usually used xhigh. The experience of relying on codex\u2019s outstandingly thorough coding earlier in the year has evaporated for me. I\u2019m seeing incredibly stupid implementations intermittently, and have simply switched to Claude until openai takes the issue seriously. As far as i could tell they haven\u2019t taken it seriously for the several ", "context": "", "use_case": "coding", "switched_from": "", "story": "GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance"}, {"id": 780728640, "item_id": 48795540, "url": "https://news.ycombinator.com/item?id=48795540", "by": "m3h", "date": "2026-07-05", "stance": "warn", "quote": "it looks like my work has suffered from the clustering issue as well", "excerpt": "Indeed, it looks like my work has suffered from the clustering issue as well: reasoning_output_tokens count percent \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 \u2501\u2501\u2501\u2501\u2501\u2501\u2501 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 0 873 28.5948 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 8 64 2.0963 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 9 60 1.9653 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 11 54 1.7688 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 516 48 1.5722 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500", "context": "", "use_case": "coding", "switched_from": "", "story": "GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance"}]}, {"label": "Hallucinations", "summary": "One commenter ranks older Codex versions in a tier that produces hallucinated output.", "quotes": [{"id": 785947720, "item_id": 49121732, "url": "https://news.ycombinator.com/item?id=49121732", "by": "user43928", "date": "2026-07-31", "stance": "warn", "quote": "\"I love hallucinated garbage\": Sonnet, Qwen 3.6, GPT 5.4 mini, GPT 5.3 Codex", "excerpt": "In my mind there are three tiers: The SOTA: Fable, GPT 5.6 Sol, Opus 5 The \"enterprise admin did not turn on the new models\": Opus 4.8, GPT 5.5 The \"I love hallucinated garbage\": Sonnet, Qwen 3.6, GPT 5.4 mini, GPT 5.3 Codex, etc. Results vary widely", "context": "", "use_case": "", "switched_from": "", "story": "Google fixed more Chrome bugs in June than over the past two years, thanks to AI"}]}, {"label": "Benchmarks vs reality", "summary": "One user found strong benchmark results did not translate into useful code.", "quotes": [{"id": 751751521, "item_id": 46984470, "url": "https://news.ycombinator.com/item?id=46984470", "by": "muzani", "date": "2026-02-11", "stance": "warn", "quote": "gpt-5.3-codex benchmarked better, but I'm not seeing this translating to useful code. It failed with the first few requests I gave it.", "excerpt": "By \"here\", certainly you don't mean HN? HN feels muted and honestly, I haven't been on here because it's quite stale by the time it gets here, and people have just been vocal about not getting it. > AI.com was sold for $70M This news is everywhere elsewhere, and it's still going on in my WhatsApp groups. I can't even find the active HN thread. This was more because people were surprised about the actual owner of ai.c", "context": "", "use_case": "coding", "switched_from": "", "story": "Ask HN: Why is everyone here so AI-hyped?"}]}, {"label": "Ignores instructions", "summary": "One user saw it repeatedly revert their manual edits despite being told to just commit.", "quotes": [{"id": 751109730, "item_id": 46944358, "url": "https://news.ycombinator.com/item?id=46944358", "by": "theorchid", "date": "2026-02-09", "stance": "warn", "quote": "the model edited the file again and added the value I deleted to the array.", "excerpt": "In my opinion, Opus 4.6 handles instructions (especially negative rules) better than Opus 4.5 and Sonnet 4.5. It also applies skills better than other models. But I think this is individual. Another example. I asked gpt-5.2-codex to add an array of 5 values to the script and write a small piece of code. Then I manually deleted one of the values in the array and asked the agent to commit. But the model edited the file", "context": "", "use_case": "agentic coding", "switched_from": "", "story": "Ask HN: Opus 4.6 ignoring instructions, how to use 4.5 in Claude Code instead?"}]}, {"label": "Slow sessions", "summary": "Described as high quality but very slow for long interactive coding sessions.", "quotes": [{"id": 733714083, "item_id": 45857130, "url": "https://news.ycombinator.com/item?id=45857130", "by": "sheepscreek", "date": "2025-11-08", "stance": "warn", "quote": "I found codex to be very high quality but so godawful slow for long interactive coding sessions.", "excerpt": "I used their $50 plan and with the previously offered Qwen3 coder 480B. While fast - none of the \u201csupported\u201d tools I tried were able to use it in a way that didn\u2019t hit the per minute request limit in a few seconds. It was incredibly frustrating. For the record, I tried OpenCoder, VSCode, Quen Coder CLI, octofriend and a few others I don\u2019t remember. Fast forward to now, when GLM 4.6 has replaced Qwen3 coder in their s", "context": "", "use_case": "coding", "switched_from": "", "story": "Cerebras Code now supports GLM 4.6 at 1000 tokens/sec"}]}], "who": [{"by": "ifwinterco", "context": "works under a monthly spend cap at work", "url": "https://news.ycombinator.com/item?id=48739025", "date": "2026-06-30"}, {"by": "polyglotfacto", "context": "experimenting with AI-assisted coding on a Servo branch", "url": "https://news.ycombinator.com/item?id=47275643", "date": "2026-03-06"}], "tags": ["for budget-capped teams", "for large codebases", "for code review", "for debugging", "for reverse engineering"], "use_cases": ["coding", "vision tasks", "main coding workhorse model", "scripting, greenfield and feature work", "peer-reviewing", "straightforward coding tasks"], "compared": [{"slug": "claude-opus", "name": "Claude Opus", "co": 13, "switched_from": 0, "switched_to": 0}, {"slug": "gpt-5", "name": "GPT-5", "co": 9, "switched_from": 0, "switched_to": 0}, {"slug": "claude", "name": "Claude", "co": 3, "switched_from": 0, "switched_to": 1}, {"slug": "cursor", "name": "Cursor", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "claude-sonnet", "name": "Claude Sonnet", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "codex", "name": "Codex", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "gemini", "name": "Gemini", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "claude-code", "name": "Claude Code", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [{"id": 753834896, "item_id": 47114681, "url": "https://news.ycombinator.com/item?id=47114681", "by": "stared", "date": "2026-02-22", "stance": "recommend", "quote": "I used GPT 5.2 Codex in Cursor for decompiling an old game and it worked (way better than Claude Code with Opus 4.5).", "excerpt": "To be honest, it is also our surprise. I mean, I used GPT 5.2 Codex in Cursor for decompiling an old game and it worked (way better than Claude Code with Opus 4.5). We tested for Opus 4.6, but waiting for public API to test on GPT 5.3 Codex. At the same time, various task can be different, and now all things that work the best end-to-end are the same as ones that are good for a typical, interactive workflow. We used ", "context": "", "use_case": "decompiling an old game", "switched_from": "", "story": "We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them"}], "threads": [{"url": "https://news.ycombinator.com/item?id=48789428", "title": "GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=47111440", "title": "We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49214008", "title": "DeepSeek V4 Flash 0731", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49120097", "title": "Google fixed more Chrome bugs in June than over the past two years, thanks to AI", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48736605", "title": "Claude Sonnet 5", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48600167", "title": "GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2", "mentions": 1}], "sentiment_by_month": {"2025-11": {"warn": 1, "neutral": 2, "recommend": 1}, "2025-12": {}, "2026-01": {"recommend": 1}, "2026-02": {"warn": 2}, "2026-03": {"recommend": 1, "neutral": 1}, "2026-04": {"neutral": 1}, "2026-05": {"recommend": 1}, "2026-06": {"recommend": 3}, "2026-07": {"warn": 3}, "2026-08": {"neutral": 1}, "2026-09": {}, "2026-10": {}}, "quotes": [{"id": 787474370, "item_id": 49217148, "url": "https://news.ycombinator.com/item?id=49217148", "by": "dools", "date": "2026-08-07", "stance": "neutral", "quote": "When I need vision capabilities I use GPT 5.3 codex", "excerpt": "I have been using deepseek v4 pro almost exclusively. I was using Kimi a lot but it just nose dived. The decline started with the release of 2.7 and accelerated with the release of 3. When I need vision capabilities I use GPT 5.3 codex and if deepseek can\u2019t figure something out after a few goes I switch to GTP 5.5 or 5.6 (I\u2019ve been giving Terra first bite recently and it does pretty well, and have used Sol a couple o", "context": "", "use_case": "vision tasks", "switched_from": "", "story": "DeepSeek V4 Flash 0731"}, {"id": 785947720, "item_id": 49121732, "url": "https://news.ycombinator.com/item?id=49121732", "by": "user43928", "date": "2026-07-31", "stance": "warn", "quote": "\"I love hallucinated garbage\": Sonnet, Qwen 3.6, GPT 5.4 mini, GPT 5.3 Codex", "excerpt": "In my mind there are three tiers: The SOTA: Fable, GPT 5.6 Sol, Opus 5 The \"enterprise admin did not turn on the new models\": Opus 4.8, GPT 5.5 The \"I love hallucinated garbage\": Sonnet, Qwen 3.6, GPT 5.4 mini, GPT 5.3 Codex, etc. Results vary widely", "context": "", "use_case": "", "switched_from": "", "story": "Google fixed more Chrome bugs in June than over the past two years, thanks to AI"}, {"id": 780728640, "item_id": 48795540, "url": "https://news.ycombinator.com/item?id=48795540", "by": "m3h", "date": "2026-07-05", "stance": "warn", "quote": "it looks like my work has suffered from the clustering issue as well", "excerpt": "Indeed, it looks like my work has suffered from the clustering issue as well: reasoning_output_tokens count percent \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 \u2501\u2501\u2501\u2501\u2501\u2501\u2501 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 0 873 28.5948 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 8 64 2.0963 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 9 60 1.9653 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 11 54 1.7688 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500 516 48 1.5722 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500", "context": "", "use_case": "coding", "switched_from": "", "story": "GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance"}, {"id": 780637536, "item_id": 48789846, "url": "https://news.ycombinator.com/item?id=48789846", "by": "zenapollo", "date": "2026-07-04", "stance": "warn", "quote": "I\u2019m seeing incredibly stupid implementations intermittently", "excerpt": "I\u2019ve definitely experienced step jumps down in quality on an almost daily basis. I usually used xhigh. The experience of relying on codex\u2019s outstandingly thorough coding earlier in the year has evaporated for me. I\u2019m seeing incredibly stupid implementations intermittently, and have simply switched to Claude until openai takes the issue seriously. As far as i could tell they haven\u2019t taken it seriously for the several ", "context": "", "use_case": "coding", "switched_from": "", "story": "GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance"}, {"id": 779824402, "item_id": 48739025, "url": "https://news.ycombinator.com/item?id=48739025", "by": "ifwinterco", "date": "2026-06-30", "stance": "recommend", "quote": "5.3-codex might be the best overall coding model OpenAI have ever released. It's 90% as good as 5.5 and costs about 20% as much", "excerpt": "I prefer GPT 5.5 to Opus but both are absurdly expensive token hogs, I can't afford to use either as my main model at $work with the monthly spend cap we have. I use Composer (since we use Cursor) or GPT 5.3-codex as my workhorse models and only break out the big guns when I have a genuinely difficult problem to solve. IMO somewhat weirdly 5.3-codex might be the best overall coding model OpenAI have ever released. It", "context": "works under a monthly spend cap at work", "use_case": "main coding workhorse model", "switched_from": "", "story": "Claude Sonnet 5"}, {"id": 777731424, "item_id": 48608214, "url": "https://news.ycombinator.com/item?id=48608214", "by": "wiether", "date": "2026-06-20", "stance": "recommend", "quote": "Since Codex-5.3 came out it was my daily driver for everything", "excerpt": "Since Codex-5.3 came out it was my daily driver for everything: quick scripting, greenfield projects, new features on old projects... Idk if it was the harness (OpenCode), my AGENT or my prompts, but I was getting exactly what I wanted, and quickly. With GPT-5.5 it tries to play smart, takes much more times and is often stuck on basic stuff that DeepSeek solves oneshot.", "context": "", "use_case": "scripting, greenfield and feature work", "switched_from": "", "story": "GPT-5.5 hallucinates 3x more than MIT-licensed GLM-5.2"}, {"id": 777144464, "item_id": 48571529, "url": "https://news.ycombinator.com/item?id=48571529", "by": "vunderba", "date": "2026-06-17", "stance": "recommend", "quote": "gpt-5.5 codex xhigh reasoning seemed to catch more edge cases and went \"deeper\" into analysis than Opus 4.7/4.8.", "excerpt": "I use LLMs more in the context of peer-reviewing and also came to a similar conclusion, gpt-5.5 codex xhigh reasoning seemed to catch more edge cases and went \"deeper\" into analysis than Opus 4.7/4.8. My preliminary tests of Fable were pretty promising but that's DOA for everyone for now.", "context": "", "use_case": "peer-reviewing", "switched_from": "", "story": "Leaked OpenAI financials show $38.5B loss and compute burn"}, {"id": 770924432, "item_id": 48182777, "url": "https://news.ycombinator.com/item?id=48182777", "by": "dgunay", "date": "2026-05-18", "stance": "recommend", "quote": "For straightforward coding tasks I use gpt-5.3-codex on high or xhigh.", "excerpt": "For straightforward coding tasks I use gpt-5.3-codex on high or xhigh. Sometimes I try 5.5 but overall 5.3-codex is more than capable enough for most of my needs and quite a bit cheaper. For more interactive/discussion/planning or orchestration stuff, I find myself going back and forth between Opus 4.7 and GPT 5.5. Still not sure which one I prefer.", "context": "", "use_case": "straightforward coding tasks", "switched_from": "", "story": "Ask HN: What LLM models are you using and why?"}, {"id": 765528193, "item_id": 47845512, "url": "https://news.ycombinator.com/item?id=47845512", "by": "0xbadcafebee", "date": "2026-04-21", "stance": "neutral", "quote": "I use GPT-5.4, GPT-5.3-Codex, Gemini-3-Pro, Opus, Sonnet, at work every week.", "excerpt": "A model cannot read your mind. It can guess, and those guesses are more likely to be wrong if you don't give it the right input, and model performance gets worse if not steered/curated properly. The output depends on the input. https://medium.com/@adambaitch/the-model-vs-the-harness-which-actually-matters-more-59dd3116bb31 | https://aakashgupta.medium.com/2025-was-agents-2026-is-agent-harnesses-heres-why-that-changes", "context": "", "use_case": "work coding", "switched_from": "", "story": "Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving"}, {"id": 757095523, "item_id": 47318470, "url": "https://news.ycombinator.com/item?id=47318470", "by": "radio879", "date": "2026-03-09", "stance": "neutral", "quote": "gpt 5.4 in there somewhere along with 5.3 codex", "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": 756410288, "item_id": 47275643, "url": "https://news.ycombinator.com/item?id=47275643", "by": "polyglotfacto", "date": "2026-03-06", "stance": "recommend", "quote": "Also switched to a more expensive coding model (GPT 5.3 Codex), which was an improvement", "excerpt": "I am in the third week of my experiment with AI-assisted coding on a personal branch of Servo (which still bans all AI contributions). First, the least relevant metric: LOC. Almost doubled from 7k to 12k in a week. Current WPT pass rate, compared to the default integration of rustnn (pywebnn) is at: https://gist.github.com/gterzian/95371d56468390177982c068480db95d This week was the grunt work: grinding through the lo", "context": "experimenting with AI-assisted coding on a Servo branch", "use_case": "AI-assisted coding", "switched_from": "", "story": null}, {"id": 753817987, "item_id": 47113624, "url": "https://news.ycombinator.com/item?id=47113624", "by": "stared", "date": "2026-02-22", "stance": "recommend", "quote": "GPT-5.2-Codex worked with it well!", "excerpt": "Thanks for sharing! It seems to be an active space, vide a recent MCP server (https://news.ycombinator.com/item?id=46882389). I you haven't tried, recommend a lot posting it as Show HN. I tried a few approaches - https://github.com/jtang613/GhidrAssistMCP (was the harderst to set) Ghidra analyzeHeadless (GPT-5.2-Codex worked with it well!) and PyGhidra (my go-to). Did you try to see which works the best? I mean, very", "context": "", "use_case": "driving Ghidra analysis", "switched_from": "", "story": "We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them"}, {"id": 751751521, "item_id": 46984470, "url": "https://news.ycombinator.com/item?id=46984470", "by": "muzani", "date": "2026-02-11", "stance": "warn", "quote": "gpt-5.3-codex benchmarked better, but I'm not seeing this translating to useful code. It failed with the first few requests I gave it.", "excerpt": "By \"here\", certainly you don't mean HN? HN feels muted and honestly, I haven't been on here because it's quite stale by the time it gets here, and people have just been vocal about not getting it. > AI.com was sold for $70M This news is everywhere elsewhere, and it's still going on in my WhatsApp groups. I can't even find the active HN thread. This was more because people were surprised about the actual owner of ai.c", "context": "", "use_case": "coding", "switched_from": "", "story": "Ask HN: Why is everyone here so AI-hyped?"}, {"id": 751109730, "item_id": 46944358, "url": "https://news.ycombinator.com/item?id=46944358", "by": "theorchid", "date": "2026-02-09", "stance": "warn", "quote": "the model edited the file again and added the value I deleted to the array.", "excerpt": "In my opinion, Opus 4.6 handles instructions (especially negative rules) better than Opus 4.5 and Sonnet 4.5. It also applies skills better than other models. But I think this is individual. Another example. I asked gpt-5.2-codex to add an array of 5 values to the script and write a small piece of code. Then I manually deleted one of the values in the array and asked the agent to commit. But the model edited the file", "context": "", "use_case": "agentic coding", "switched_from": "", "story": "Ask HN: Opus 4.6 ignoring instructions, how to use 4.5 in Claude Code instead?"}, {"id": 747416851, "item_id": 46713553, "url": "https://news.ycombinator.com/item?id=46713553", "by": "Octokat", "date": "2026-01-21", "stance": "recommend", "quote": "the other 2 do a much better job at the level and complexity of large codebases", "excerpt": "I use Cursor with a max subscription through work and using Gemini in multi-model mode alongside Opus 4.5 and GPT 5.2 Codex, Gemini's been the worst performer. Not saying it's a bad model just that the other 2 do a much better job at the level and complexity of large codebases. Just my experience.", "context": "", "use_case": "large codebase coding", "switched_from": "", "story": "Ask HN: What single AI tool/technique 10x'd your productivity last year?"}, {"id": 736670641, "item_id": 46041915, "url": "https://news.ycombinator.com/item?id=46041915", "by": "nickandbro", "date": "2025-11-24", "stance": "recommend", "quote": "5.1 Codex I use for a narrowly defined task where I can just fire and forget it.", "excerpt": "I use the following models like so nowadays: Gemini is great, when you have gitingested the code of pypi package and want to use it as context. This comes in handy for tasks and repos outside the model's training data. 5.1 Codex I use for a narrowly defined task where I can just fire and forget it. For example, codex will troubleshoot why a websocket is not working, by running its own curl requests within cursor or e", "context": "", "use_case": "fire-and-forget debugging tasks", "switched_from": "", "story": "Claude Opus 4.5"}, {"id": 736668881, "item_id": 46041805, "url": "https://news.ycombinator.com/item?id=46041805", "by": "guiprav", "date": "2025-11-24", "stance": "neutral", "quote": "the voice assistant is able to call GPT 5.1 Codex to generate beautiful HTML elements", "excerpt": "https://vimeo.com/1140238012?share=copy&fl=sv&fe=ci So this is my visual app builder Webfoundry, I used it to create a handful web apps including MEATEOR, the P2P Grindr alternative I posted about here: https://news.ycombinator.com/item?id=46040436 This demo shows how the voice assistant is able to call GPT 5.1 Codex to generate beautiful HTML elements for your page on-demand based on voice specs. Other than that, We", "context": "", "use_case": "HTML generation from voice specs", "switched_from": "", "story": "Webfoundry gets GPT 5.1 Codex HTML generation through voice assistant"}, {"id": 733945314, "item_id": 45871582, "url": "https://news.ycombinator.com/item?id=45871582", "by": "wasi0013", "date": "2025-11-09", "stance": "neutral", "quote": "where it failed, switched to the sonnet 4.5 or GPT 5  codex", "excerpt": "Experimenting with various AI models via GitHub Co-Pilot on an extremely niche project to see how far these models have progressed. Used like ~60% of the premium quota to develop the following projects: website: https://murajah.pages.dev/ Play Store: https://play.google.com/store/apps/details?id=com.murajah.webview Manual audio splitting tool for the above project: audio-splitter-6b3.pages.dev/ I've always been skept", "context": "", "use_case": "debugging when Haiku failed", "switched_from": "", "story": "Ask HN: What Are You Working On? (Nov 2025)"}, {"id": 733714083, "item_id": 45857130, "url": "https://news.ycombinator.com/item?id=45857130", "by": "sheepscreek", "date": "2025-11-08", "stance": "warn", "quote": "I found codex to be very high quality but so godawful slow for long interactive coding sessions.", "excerpt": "I used their $50 plan and with the previously offered Qwen3 coder 480B. While fast - none of the \u201csupported\u201d tools I tried were able to use it in a way that didn\u2019t hit the per minute request limit in a few seconds. It was incredibly frustrating. For the record, I tried OpenCoder, VSCode, Quen Coder CLI, octofriend and a few others I don\u2019t remember. Fast forward to now, when GLM 4.6 has replaced Qwen3 coder in their s", "context": "", "use_case": "coding", "switched_from": "", "story": "Cerebras Code now supports GLM 4.6 at 1000 tokens/sec"}]}