OpenAI Releases GPT-6.2 Astra and GPT-6.2 Sol With Cheaper API Pricing
OpenAI released the AI models GPT-6.2 Astra and GPT-6.2 Sol on November 17 US time, offering both through ChatGPT Work, Codex, and an API.
3 sources 16h ago Updated
Reporting from 3 sources: ASCII.jp, GameBusiness.jp, GIGAZINE.
Why it matters
OpenAI cut Astra's input and output price by 20% while its own tests put the model behind Anthropic's Claude Fable 5.2 on the Artificial Analysis Intelligence Index, so the release competes on cost per finished task rather than on top-line benchmark scores.
Key Facts
- OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17 US time through ChatGPT Work, Codex, and an API.
- GPT-6.2 Astra costs $8 per million input tokens, $0.40 cached input, and $40 output, 20% cheaper on input and output than GPT-6 Astra.
- GPT-6.2 Sol costs $2 per million input tokens and $10 output, the same as GPT-6.1 Sol.
- GPT-6.2 Astra scored 5.2 points above GPT-6 Astra on the Agents' Last Exam test at less than one-third the cost per task.
- GPT-6.2 Astra ranked second on the Artificial Analysis Intelligence Index, behind Anthropic's Claude Fable 5.2.
Plus, Pro, Enterprise, and Edu subscribers get access; Enterprise and Edu workspaces need an administrator to enable them. Astra is the top-tier model for complex work, programming, and scientific research, and OpenAI says it produces slides, spreadsheets, and documents that follow supplied templates and industry conventions. On the Agents' Last Exam test of multi-step professional work, Astra scored 5.2 points above GPT-6 Astra at less than one-third the cost per task. Sol targets everyday work such as writing and sequential research tasks; on FrontierCode 2.0 it came within 0.5 points of Astra at about one-fifth the cost. Astra's API price is $8 per million input tokens, $0.40 cached input, and $40 output, 20% cheaper on input and output than GPT-6 Astra. Sol stays at $2 input and $10 output.
OpenAI frames the release as a step down in token use, not only in price. On its internal GDPval v2 test, built from real consulting and finance projects, Astra scored 6.2 points above GPT-6 Astra while using 22% fewer output tokens.
Independent numbers are less flattering. On the Artificial Analysis Intelligence Index, Astra placed second, behind Anthropic's Claude Fable 5.2. On the Artificial Analysis Coding Agent Index it matched GPT-6 Astra, while Sol came in 3 points below GPT-6.1 Sol.
Other results from the announcement:
- Terminal-Bench-Science 0.1: Astra beat GPT-6 Astra by 9 percentage points.
- Agents' Last Exam: Astra also beat Claude Fable 5.2 by 5.5 points.
- PacBench: Astra scored 92 in 6.6 minutes at $2.04; Sol scored 87 in 6.3 minutes at $0.49. Kimi K3.1 Pro scored 93 but took 11.5 minutes.
- LifeVal: Sol exceeded GPT-6 Astra at about one-seventh the cost.
- Plotly data analysis test: Sol answered 40 of 43 questions correctly at $1.22, down from GPT-6.1 Sol's $2.61, with input tokens falling from 7.8 million to 4.6 million.
Both models reached OpenAI's "Critical" threshold for cybersecurity, and restrictions on high-risk requests were tightened. Some legitimate vulnerability research can be restricted as a result. Astra also hit "High" on OpenAI's evaluation of self-improvement capability, which OpenAI describes as a first for an externally released model.
Demos include a 3D car model in Blender with animated suspension, a playable race game, and a Colosseum viewable from any angle. As of publication, plans to offer the models in regular ChatGPT had not been clarified.
Written with AI by Yomimono from the 3 cited sources below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google