OpenAI releases GPT-6.2 Astra and Sol
OpenAI is releasing GPT-6.2 Astra and GPT-6.2 Sol, describing Astra as its frontier model for difficult projects and Sol as an efficient model for everyday work. Both are available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex.

Key takeaways · 4
- 01
Both models are available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex.
- 02
OpenAI’s quoted API prices are $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens.
- 03
Astra’s coding benchmark score did not improve over GPT-6 Astra’s, while Sol lost three points against GPT-6.1 Sol.
- 04
OpenAI says both models meet its Critical cybersecurity capability threshold and that it tightened restrictions on higher-risk assistance.
Two models for different workloads
OpenAI is releasing GPT-6.2 Astra and GPT-6.2 Sol.[1] The company describes Astra as its frontier model for difficult projects, while it positions Sol as an efficient option for everyday work.[2] Both models are available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex.[3] That gives eligible teams access through those products, while API pricing is a separate consideration for teams evaluating model use through the API. OpenAI’s descriptions distinguish the intended roles of the models; they do not establish that one will perform better for every team or task.
Pricing and benchmark position
OpenAI’s quoted API prices are $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens.[2] OpenAI says Astra’s API input and output prices are 20% below GPT-6 Astra’s, with a 60% discount on cached input; Sol’s API token prices remain at GPT-6.1 Sol’s rates.[2] Artificial Analysis ranked Astra second in its Intelligence Index, with 62 points to Claude Fable 5.2’s 65, and reported a five-point gain over GPT-6 Astra.[4][3] The index combines nine evaluations covering knowledge work, coding, science and general knowledge.[2]
Coding results are mixed
The Coding Agent Index did not show an improvement for Astra: Artificial Analysis found its score unchanged from GPT-6 Astra’s.[4] Sol scored three points lower than GPT-6.1 Sol.[4] The evaluator measured this index using the Codex harness at maximum effort.[4] It attributed part of Sol’s decline to a 14-point drop on SWE-Atlas-QnA-AA, where Sol more often omitted key facts than its predecessor.[4] Artificial Analysis also said OpenAI models occupied most of the cost-per-task Pareto frontier alongside models from Anthropic and Meta.[4] Teams should treat these as benchmark findings, not a guarantee of performance on their own coding tasks.
Safety thresholds and evaluation caveats
OpenAI classified both models as meeting its Critical threshold for cybersecurity capabilities and said it tightened default restrictions on higher-risk cybersecurity assistance, adding vetting for more permissive access.[1] OpenAI also said Astra met its High threshold for AI self-improvement capabilities, while Sol fell below it.[1] On Artificial Analysis’s AA-Omniscience measure, maximum-effort hallucination rates were 59% for Astra and 55% for Sol.[4] The questions in that measure are unusually hard, and the reported numbers do not reflect everyday hallucination rates.[3] These distinctions matter when interpreting both safety disclosures and benchmark results.
The launch gives eligible teams another choice between a model positioned for difficult projects and one positioned for everyday work. Benchmark results suggest teams should test their own coding tasks and weigh API costs alongside capability and access controls.
Why it matters
AI training for your team
kju.ai is built for teams: daily sessions, team analytics, and rollout support for your organization.
How this developed
17 November 2026
OpenAI releases GPT-6.2 Astra and Sol
Sources
- GPT-6.2 Astra and Sol system carddeploymentsafety.openai.com
- Introducing GPT-6.2 Astra and Sol | OpenAIopenai.com
- OpenAI launches GPT-6.2 Astra and Sol with lower API prices and some benchmark gainsthe-decoder.com
- OpenAI has the #2 model in the Intelligence Index with the release of GPT-6.2 Astra and GPT-6.2 Sol | Artificial Analysisartificialanalysis.ai