0.79
- 开始
- 2024年7月18日可信度 95%来自help.openai.com
GPT-4o mini Released
OpenAI introduces GPT-4o mini, its most capable and cost-efficient small model, surpassing GPT-3.5 Turbo on academic benchmarks.
- OpenAI introduced GPT-4o mini on July 18, 2024 as its most cost-efficient small model[1][2]
- Available in the Assistants, Chat Completions and Batch APIs; in ChatGPT Free, Plus and Team users get it that day in place of GPT-3.5, Enterprise the following week[2]
- First model in the API to apply the instruction hierarchy method against jailbreaks and prompt injections[2]
- OpenAI's release notes list it as the July 18, 2024 entry, later replaced in ChatGPT by GPT-4.1 mini (May 14, 2025)[1]
Notable features
- A small model that surpasses GPT-3.5 Turbo and other small models on textual and multimodal academic benchmarks, supports the same languages as GPT-4o and shares its improved tokenizer, with stronger function calling and long-context performance than GPT-3.5 Turbo[3]
- It scores 82% on MMLU (Gemini Flash 77.9%, Claude Haiku 73.8%) and 87.2% on HumanEval, and outperforms GPT-4 on chat preferences in the LMSYS leaderboard[2]
- It scores 87.0% on MGSM math reasoning (Gemini Flash 75.5%, Claude Haiku 71.7%) and 59.4% on the MMMU multimodal reasoning eval (56.1% and 50.2%)[3]
- Priced at 15 cents per million input tokens and 60 cents per million output tokens, more than 60% cheaper than GPT-3.5 Turbo[2]
- Context window of 128K tokens, up to 16K output tokens per request, knowledge to October 2023; text and vision in the API with other modalities to come[2]
参考资料 3可信度 90%
第一项始终是图钉的来源。总体可信度是各份资料对上方所用开始和结束时间支持程度的加权平均；资料每比最新的一份旧 180 天，权重减半。
- [1]95%OpenAI Model Release Notes: Introducing GPT-4o mini (July 18, 2024)help.openai.com· 发布于 2026年9月19日· 开始 2024年7月18日· 占评分 49%
OpenAI's[2] own model release notes date this entry July 18, 2024
- [2]85%GPT-4o mini: advancing cost-efficient intelligenceopenai.com· 添加于 2026年9月19日· 开始 2024年7月18日· 占评分 49%
OpenAI's[1] announcement post, linked from the release note entry.
- [3]88%GPT-4o mini: advancing cost-efficient intelligence | OpenAI (Wayback Machine copy)web.archive.org· 发表于 2024年7月18日· 占评分 2%
Archived copy of OpenAI's[1][2] GPT-4o mini announcement, which says "On MGSM, measuring math reasoning, GPT-4o mini scored 87.0%" and "scoring 59.4%" on MMMU, and that it shares GPT-4o's improved tokenizer.
建议更正
有遗漏或错误吗？用你自己的话说明：能佐证此图钉的链接、不同的开始或结束日期及理由，或缺失、有误的信息。AI 会对照此图钉的来源进行核实，搜索更好的来源，并添加任何支持你说法的页面。图钉自身的来源仍然最重要。
