What is GPT-6.2 Astra?
GPT-6.2 Astra is OpenAI's new frontier model at $8/$40 per million tokens with a 1M context window. See benchmarks, availability and how Sol compares.
Updated ·
GPT-6.2 Astra: Pricing, 1M Context, Benchmarks and Availability
Announcement date:
GPT-6.2 Astra is one of two models OpenAI shipped together, alongside GPT-6.2 Sol. Astra is the higher-priced, higher-scoring option: $8 per million input tokens, $0.40 cached input and $40 output, with a 1M context window and a 128K output limit. It is rolling out to ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu tiers, and to developers through the API as gpt-6.2-astra. Sol keeps its predecessor's pricing at $2, $0.10 and $10. Benchmark results are mixed: Astra trails Fable 5.2 on the Intelligence Index but leads it on several agentic and 3D evaluations, and OpenAI rated Astra High for AI Self-Improvement — a first for one of its externally released models.

What GPT-6.2 Astra is
GPT-6.2 Astra is a new OpenAI model released together with GPT-6.2 Sol. The two form a pair: Astra is positioned as the stronger but more expensive model, while Sol keeps the pricing of its predecessor and is described in coverage as the likely main workhorse.
The launch came two days before Anthropic's expected first trading day, a timing detail worth noting for anyone tracking competitive releases rather than model capability itself.
Astra is not a single configuration. Beyond the base model, there is an Enterprise configuration called Astra for Law that adds a legal search index.
- Released alongside GPT-6.2 Sol as a two-model launch
- Astra is the pricier, higher-scoring sibling; Sol holds prior pricing
- An Enterprise variant, Astra for Law, adds a legal search index
Pricing, context window and output limit
Astra is priced at $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens. Sol, by contrast, keeps its predecessor's pricing at $2 input, $0.10 cached input and $10 output.
Both models share the same context and output ceilings: a 1M context window and a 128K output limit. That means the choice between them is mainly about cost per task versus capability, not about how much text they can hold.
Cost per task is where the picture gets more interesting than list prices suggest. On Perplexity's WANDR, Astra scored 0.783 at about $12.54 per task while Sol reached 0.746 at about $2.65. In Artificial Analysis tests, Sol ties September's Astra at roughly 15% of the cost per task.
- Astra: $8 input / $0.40 cached input / $40 output per million tokens
- Sol: $2 input / $0.10 cached input / $10 output per million tokens
- Both: 1M context window and 128K output limit
- WANDR cost per task: about $12.54 for Astra versus about $2.65 for Sol
Benchmark results and how they compare with Fable 5.2
The headline capability number is mixed. New Astra scores 62 on the Intelligence Index, just behind Fable 5.2 at 65. On agentic and spatial evaluations, however, Astra often leads.
Mercor's OpenWorldEngineer puts Astra at 67.2%, Fable 5.2 at 54.6% and GPT-6 Astra at 44.2%. Astra also led AA-3D with 1,318, ahead of Fable at 1,271 and Sol at 1,259, and scored 79 on Artificial Analysis's robotics preview against Sol's 78 and Fable's 75.
OpenAI's own reporting adds efficiency claims: on Agents' Last Exam 1.1, Astra at xhigh gained 5.2 points over GPT-6 Astra at max while using less than a third of the cost per task, and on GDPval v2 Astra used 22% fewer output tokens than its predecessor. These are vendor-reported figures and should be read as such.
- Intelligence Index: Astra 62, Fable 5.2 65
- OpenWorldEngineer: Astra 67.2%, Fable 5.2 54.6%, GPT-6 Astra 44.2%
- AA-3D: Astra 1,318, Fable 1,271, Sol 1,259
- Robotics preview: Astra 79, Sol 78, Fable 75
- Agents' Last Exam 1.1: +5.2 points over GPT-6 Astra at max, under a third of cost per task (OpenAI-reported)
- GDPval v2: 22% fewer output tokens than predecessor (OpenAI-reported)
Astra for Law and the Harvey LAB-AA result
Astra for Law is an Enterprise configuration rather than a separate public model. It adds a legal search index on top of the base model, aimed at legal research and document work.
The measured difference is substantial on one legal benchmark: Artificial Analysis recorded 20.8% on Harvey LAB-AA for Astra for Law, versus 15.0% for base Astra. That is a meaningful relative gain, though the absolute scores are low enough that neither should be treated as a finished legal reasoning system.
- Enterprise-only configuration with an added legal search index
- Harvey LAB-AA: 20.8% for Astra for Law versus 15.0% for base Astra
- Absolute scores remain low; treat as a research-grade signal, not a guarantee
Where you can use it
Astra is rolling out to ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu tiers. Developers can reach it through the API under the model name gpt-6.2-astra; Sol is available as gpt-6.2-sol.
The source does not state regional restrictions, rate limits, or whether free-tier users get any access, so those details should be treated as unknown until OpenAI documents them.
If you are choosing between the two, a practical approach is to start with Sol for high-volume or cost-sensitive work and reserve Astra for tasks where the agentic benchmark gap matters — long multi-step engineering, spatial reasoning, or legal research with the Enterprise configuration.
- ChatGPT Work and Codex: Plus, Pro, Enterprise and Edu tiers
- API model names: gpt-6.2-astra and gpt-6.2-sol
- Regional availability, rate limits and free-tier access are not stated in the source
The High AI Self-Improvement rating
OpenAI rated Astra High for AI Self-Improvement, which the source describes as a first for one of its externally released models. The rating concerns the model's potential to contribute to improving AI systems themselves.
The source does not detail the evaluation methodology behind the rating or what mitigations accompany it. Organizations with strict internal AI governance should factor the rating into procurement decisions and check OpenAI's own system card for the underlying reasoning.
- Rated High for AI Self-Improvement — a first for an externally released OpenAI model
- Methodology and mitigations are not described in the source
- Worth reviewing in OpenAI's system card before enterprise deployment
Questions
How much does GPT-6.2 Astra cost per million tokens, and how does it compare with GPT-6.2 Sol?
Astra costs $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens. Sol keeps its predecessor's pricing at $2 input, $0.10 cached input and $10 output. Both share a 1M context window and a 128K output limit, so the decision is mainly capability versus cost per task.
Where can I use GPT-6.2 Astra — ChatGPT Work, Codex, or the API?
All three. Astra is rolling out to ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu tiers, and developers can access it through the API as gpt-6.2-astra. Sol is available as gpt-6.2-sol. The source does not specify regional limits or free-tier access.
How does GPT-6.2 Astra compare with Fable 5.2 on benchmarks?
It depends on the benchmark. Astra scores 62 on the Intelligence Index versus Fable 5.2 at 65, but leads on Mercor's OpenWorldEngineer (67.2% versus 54.6%), AA-3D (1,318 versus 1,271) and Artificial Analysis's robotics preview (79 versus 75).
Why did OpenAI rate GPT-6.2 Astra High for AI Self-Improvement?
OpenAI assigned the High rating for AI Self-Improvement, described as a first for one of its externally released models. The source does not explain the methodology or accompanying mitigations, so the underlying reasoning should be checked in OpenAI's system card.
Original sources
Tool Info
Has this tool changed? Help us keep it up to date
GPT-6.2 Astra Reviews
Loading reviews…