OpenAI Launches GPT-6.2 Astra and Sol: Full Specs, Benchmarks and Pricing
OpenAI's new flagship matches Anthropic on list price at $8 / $40 - and shifts the contest to cost per finished task.

OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol, its latest pair of frontier models, less than a week after Anthropic cut the price of its own flagship. The launch, on Tuesday 17 November, puts the two companies on identical list pricing for their top models – and turns the contest from price per token into cost per finished task.
At a glance
- GPT-6.2 Astra is the new flagship for complex professional work, at $8 per million input tokens and $40 per million output tokens – 20% below GPT-6 Astra.
- GPT-6.2 Sol is the cheaper model for everyday multi-step work, at $2 input / $10 output per million tokens, unchanged from GPT-6.1 Sol.
- Astra matches Anthropic’s Claude Fable 5.2 on list price, at $8 / $40. OpenAI says Astra sets new highs on professional-work and scientific-research evals at less than a third of the cost per task of GPT-6 Astra. That is OpenAI’s claim; independent results have not yet been published.
- Both models are available in ChatGPT Work, Codex and the API as
gpt-6.2-astraandgpt-6.2-sol. They are Work / Codex only, not regular Chat.
Astra is the model OpenAI wants teams to reach for when the work is hard: building financial models, working through scientific problems, and running long, multi-step projects in ChatGPT Work and Codex. Sol is the everyday counterpart, built for routine multi-step work and agentic workflows at a quarter of Astra’s input price.
A price match, and a benchmark claim to test
The timing is deliberate. Anthropic launched Claude Fable 5.2 on 9 November at $8 / $40, a 20% cut from Fable 5.1, and Chain AI Wire covered that launch on 10 November. OpenAI has now matched that price – and says Astra beats Fable on a benchmark for multistep professional work. Until independent benchmarks arrive, that claim should be read as OpenAI’s, not as a settled result. Google’s Gemini 4 Carbon arrived in the same November wave of flagship launches.
What matters for users is not the headline price alone. OpenAI’s own framing is cost per task: Astra, it says, reaches new highs on professional and scientific evals at less than a third of the cost per task of GPT-6 Astra. If that holds in real use, the 20% list-price cut understates the saving. The test will be whether teams actually finish the same work with fewer tokens – and whether the rival benchmarks, when they come, agree. Pricing is listed on OpenAI’s API pricing page.
Pricing compared
| Model | Input / 1M tokens | Output / 1M tokens | Best for |
|---|---|---|---|
| GPT-6.2 Astra | $8 | $40 | Complex professional work, financial modelling, scientific research |
| GPT-6.2 Sol | $2 | $10 | Everyday multi-step work, agentic workflows |
| GPT-6 Astra (previous) | $10 | $50 | Previous flagship |
| GPT-6.1 Sol | $2 | $10 | Previous everyday model |
| Claude Fable 5.2 | $8 | $40 | Anthropic flagship, long-horizon agent work |
Availability and safety
Availability: ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users; API model IDs gpt-6.2-astra and gpt-6.2-sol. Both models are Work / Codex only, not regular Chat. An Astra for Law configuration is also available at $8 / $40, with access requiring approval through Trusted Access.
Safety note: OpenAI’s system card for the release puts both models at the company’s Critical level for cybersecurity capability, with tighter limits on higher-risk security help (as summarised 18 Nov). That is a vendor disclosure, not an independent assessment.
What is GPT-6.2 Astra?
OpenAI’s most capable model for complex professional projects, launched 17 Nov 2026.
How much does GPT-6.2 Astra cost?
$8 per million input tokens and $40 per million output tokens via the API; cached input is $0.40.
How much does GPT-6.2 Sol cost?
$2 input / $10 output per million tokens, unchanged from GPT-6.1 Sol.
Is GPT-6.2 cheaper than Claude Fable 5.2?
On list price, no – both flagships are $8 / $40. The difference, if any, will be cost per completed task.
Where can I use GPT-6.2?
In ChatGPT Work and Codex (Plus, Pro, Enterprise, Edu) and via the OpenAI API. It is not in regular Chat.
Should I switch from GPT-6 Astra?
If you run heavy professional workloads in Work or Codex, the 20% price cut and OpenAI’s claimed cost-per-task improvement make it worth testing on your own workload before switching production use.


