CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins93% CONFIDENCEOverall confidence: 93%How well the pin's source and references back up its dates.Weighted average of how firmly 3 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better
OpenAI's[1] own model release notes date this entry September 12, 2024- Start
- Sep 12, 202495% CONFIDENCEhelp.openai.com
OpenAI o1-preview and o1-mini Released
OpenAI releases o1-preview and o1-mini, the first of a new series of models that spend more time thinking before they respond.
- OpenAI released o1-preview and o1-mini on September 12, 2024, with weekly ChatGPT message limits of 30 and 50 respectively[1]
- OpenAI published a system card on September 12 for the models[2]
- An October 2024 podcast describes o1's chain-of-thought reasoning, its strengths on complex coding, head-to-head safety tests against GPT-4o, and jailbreaking, hallucination, deception and persuasion evaluations; it uses only the episode description, not a transcript[2]
- OpenAI used GPT-4o to monitor o1's behavior, and ran autonomy and red-team tests on cybersecurity and misinformation, per the podcast description[2]
Notable features
- The first of a new series of reasoning models trained to spend more time thinking through problems before they respond, learning to refine their thinking, try different strategies and recognise mistakes; OpenAI reset the counter to 1 and named the series OpenAI o1[3]
- In OpenAI's tests the next model update, then in development, scored 83% on an International Mathematics Olympiad qualifying exam against 13% for GPT-4o, and reached the 89th percentile in Codeforces contests[3]
- A new safety training approach has the model reason about safety rules in context; on one of OpenAI's hardest jailbreaking tests o1-preview scored 84 out of 100, against 22 for GPT-4o[3]
- o1-mini is a faster, cheaper reasoning model particularly effective at coding, 80% cheaper than o1-preview[3]
- At launch the models lacked web browsing and file and image uploads in ChatGPT, and the API lacked function calling, streaming and system messages, with API access for usage tier 5 developers at 20 requests per minute[3]
Buy onChatGPT