OpenAI o1-preview (2024年)（オーワンプレビュー）
2024年9月OpenAI発表o1-preview・Industry-leading reasoning chain-of-thought RLHF + Industry-leading 89.4 Codeforces percentile + Industry-leading 78.0 PhD-level GPQA + Industry-leading OpenAI o1-preview reasoning-focused LLM frontier 2024。
概要
OpenAI o1-preview は2024年9月OpenAI発表・Industry-leading reasoning chain-of-thought RLHF 2024年OpenAI o1-preview reasoning-focused LLM frontier position確立。Industry-leading 89.4 Codeforces + 78.0 GPQA + 83.3 AIME 2024 + ChatGPT Plus + API access + test-time compute scaling innovation。
主な特徴・仕組み
- Industry-leading OpenAI 2015 SF Sam Altman + Greg Brockman + Ilya Sutskever 9 years
- Industry-leading chain-of-thought RLHF internal long thinking signature
- Industry-leading 89.4 Codeforces percentile (vs GPT-4o 11)
- Industry-leading 78.0 PhD-level GPQA Diamond signature
- Industry-leading 83.3 AIME 2024 math (vs GPT-4o 13.4)
- Industry-leading scientific reasoning + STEM-focused signature
- Industry-leading hidden reasoning tokens signature
- Industry-leading ChatGPT Plus + API access ($15/$60 per 1M tokens)
- Industry-leading test-time compute scaling innovation signature
スペック比較表
| LLM Reasoning Model 2024 | Developer | Approach | Access | Industry Position |
|---|---|---|---|---|
| o1-preview | OpenAI | Chain-of-thought RLHF | ChatGPT Plus + API | Industry-leading OpenAI o1-preview frontier |
| o1-mini | OpenAI | Chain-of-thought RLHF compact | ChatGPT Plus + API | Industry-leading OpenAI o1-mini cost-efficient |
| Skywork-o1 | Kunlun Tech | o1-like open-source | MIT | Industry-leading Skywork-o1 open-source |
| QwQ | Alibaba | Reasoning 32B open-source | Apache 2.0 | Industry-leading Alibaba QwQ 32B open |
| DeepSeek-R1 Lite Preview | DeepSeek | Reasoning preview | API | Industry-leading DeepSeek R1 Lite Preview |
まとめ
OpenAI o1-preview = 2024年9月OpenAI発表のpremium reasoning-focused LLM frontier。Industry-leading chain-of-thought RLHF + 89.4 Codeforces + 78.0 PhD GPQA + 83.3 AIME + scientific reasoning + STEM + hidden reasoning tokens + ChatGPT Plus + API + test-time compute scaling + OpenAI 2015 SF Sam Altman 9 years position確立。