OpenAI 发布 GPT-6.2 双模型:Astra 与 Sol
OpenAI 一次发了两个模型:Astra 拼最强推理,Sol 用五分之一成本接近它的编码水平。做 agent 或批量代码任务的开发者可以先试 Sol,成本敏感的团队应该最感兴趣。
OpenAI 发布 GPT-6.2 系列,包含 Astra 和 Sol 两个模型。Astra 面向高难度任务,在 Agents' Last Exam 基准上以 xhigh 设置比 GPT-6 Astra(max 设置)高出 5.2 分,且单任务成本不到一半;在 GDPval v2 基准上,它在咨询、金融、设计等多步项目中超越 GPT-6 Astra,同时输出 token 减少 22%。
Sol 主打性价比,在 FrontierCode 2.0 基准上,Sol 在真实代码库的 merge-ready 代码变更任务上接近 Astra 的表现,成本约为 Astra 的五分之一。两个模型在代码质量、前端生成、按模板制作幻灯片和表格、科学推理等场景均有提升。
定价方面,Astra 为每 100 万 token 输入 $8、缓存输入 $0.40、输出 $40;Sol 为输入 $2、缓存输入 $0.10、输出 $10。两个模型即日在 ChatGPT Work 和 Codex 中上线,并通过 OpenAI API 提供。
Introducing GPT-6.2 Astra and Sol ✨
Astra for your hardest projects. Sol for strong results at a fraction of the cost.
💻 Cleaner code and better frontends 📊 Slides and spreadsheets that follow your template 🔬 Stronger reasoning on scientific tasks
> Agents’ Last Exam: 6.2 Astra at xhigh scores 5.2 points above GPT-6 Astra at max, at less than half the cost per task
> FrontierCode 2.0: Sol nearly matches Astra on merge-ready changes to real codebases at about one-fifth the cost
> GDPval v2: Astra outperforms GPT-6 Astra on multistep projects in consulting, finance, and design, using 22% fewer output tokens
Rolling out today in ChatGPT Work and Codex, and available in the OpenAI API.
API pricing per 1M tokens: > Astra: $8 input / $0.40 cached input / $40 output > Sol: $2 input / $0.10 cached input / $10 output
Excited to see what you all cook with these 🔥
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include GPT-6.2, OpenAI, Sol, reasoning models.
Source attribution: curated from Vaibhav Srivastav (@reach_vb).