Model releaseBreaking

OpenAI releases GPT-6.2 Astra and Sol, cutting Astra’s API price by 20%

Two models, one split: Astra for hard professional work at $8/$40 per million tokens, Sol for everyday coding and agents at $2/$10. Both have a 1.05-million-token context window. Most performance claims are still OpenAI’s own.

Promptea Editorial4 min read

OpenAI released two new models on November 17, 2026: GPT-6.2 Astra, aimed at demanding reasoning, coding and professional work, and GPT-6.2 Sol, aimed at everyday coding and agent workflows. Both are available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and in the API as gpt-6.2-astra and gpt-6.2-sol, according to OpenAI’s developer community announcement and its API changelog.

OpenAI’s main pitch is efficiency: similar or better results for less money per task. Most of the evidence for that comes from OpenAI’s own evaluations. The first independent results are more mixed.

What it costs

According to OpenAI’s model pages, Astra costs $8 per million input tokens and $40 per million output tokens, with cached input at $0.40. The Decoder reports that this is 20% below GPT-6 Astra. Sol costs $2 input and $10 output, with cached input at $0.10, the same as GPT-6.1 Sol.

The pricing pages include several details that matter for real bills:

  • Long prompts cost more. Above 272,000 input tokens, input and cache rates double and output is billed at 1.5x for the whole request (Astra, Sol).
  • Cache writes are billed. A 30-minute cache write costs $10 per million tokens on Astra and $2.50 on Sol. Longer 12-hour retention costs more.
  • Batch is 50% of Standard rates. Fast mode is 2x Standard. Ultrafast is 3x Standard through February 28, 2027, and rises to 6x from March 1, 2027.
  • Background windows. The changelog also announced two-hour and eight-hour background completion windows in a Responses API public beta. They are 30% and 40% cheaper than Standard, and they cover both new models.

Specs and API constraints

Both models have a 1,050,000-token context window, a maximum input of 922,000 tokens and a maximum output of 128,000 tokens. They take text and images as input and return text only. Knowledge cutoffs are May 18, 2026 for Astra and June 1, 2026 for Sol. Both support structured outputs, function calling, streaming and prompt caching. Neither supports fine-tuning or predicted outputs.

There is one integration caveat. The model pages list Chat Completions as a supported endpoint, but the changelog says tool calling requires the Responses API. OpenAI links a migration guide. Teams with agents built on Chat Completions should check this before they switch model names.

What OpenAI claims, and what others found

In its announcement, OpenAI says Astra sets new highs on professional-work and scientific evaluations at less than a third of GPT-6 Astra’s cost per task. It also says Astra at xhigh effort beats GPT-6 Astra at max effort by 5.2 points on Agents’ Last Exam, at less than half the cost per task. For Sol, OpenAI says it comes within about half a point of Astra on its FrontierCode 2.0 coding benchmark at roughly a fifth of the cost. These are vendor-reported figures.

The Decoder gathered the first outside evaluations. Its summary is mixed:

  • On the Artificial Analysis Intelligence Index, The Decoder reports Astra at 62, second behind Claude Fable 5.2 at 65, with Sol at 57. It also reports that Artificial Analysis saw no coding gain for Astra over GPT-6 Astra and a three-point drop for Sol versus GPT-6.1 Sol.
  • Epoch AI reportedly scored Astra at 174 on its Capabilities Index, tied with Fable 5.2, and found it weaker than Fable 5.2 on real-world research tasks.
  • Vals AI reportedly ranks Astra 7th and Sol 9th on its index, with low placements on legal benchmarks.

Promptea did not open those evaluators’ pages directly. The figures above are as reported by The Decoder and are early, first-day numbers.

Safety notes

OpenAI’s announcement says both models ship with tighter safeguards for higher-risk cybersecurity help, and that approved defenders could get fewer refusals through its Daybreak program. The Decoder, citing OpenAI’s safety materials, reports that OpenAI describes Astra as its first externally released model to reach the “High” threshold for AI self-improvement under its Preparedness Framework, and lists added monitoring measures. Promptea could not open OpenAI’s full launch page or system card, which are blocked to automated clients, so this point rests on that report.

What changes for prompting and model choice

The practical decision here is Sol versus Astra. Sol costs a quarter of Astra’s per-token price, and OpenAI’s own coding figure puts it close to Astra. That makes Sol a reasonable default for coding agents and routine work, with Astra kept for tasks where you can measure the gain. Independent coding results don’t yet show a clear improvement over the previous generation, so test on your own repositories before you migrate.

The 1.05-million-token window does not mean long prompts are cheap. Pricing changes above 272,000 input tokens, so retrieval and context trimming still save real money. If you use prompt caching, budget for cache-write charges, not just the discounted cached reads.

Why this matters

  • It resets OpenAI’s API price ladder: frontier-tier Astra gets 20% cheaper and Sol keeps its $2/$10 price.
  • Tool calling requires the Responses API, which affects agent stacks still on Chat Completions.
  • Early independent scores don’t clearly back the vendor’s coding claims, so model choice needs your own testing.

Key takeaways

  • GPT-6.2 Astra costs $8/$40 per million tokens and GPT-6.2 Sol costs $2/$10. Both have a 1,050,000-token context window.
  • Above 272,000 input tokens, input and cache rates double and output is billed at 1.5x.
  • Benchmark claims are mostly OpenAI’s own. Outside evaluators reported mixed results, including no coding gain for Astra.

Sources

  1. OpenAI Developer CommunityPrimary
    GPT-6.2 Astra and Sol are now available
    community.openai.com
  2. OpenAIPrimary
    API changelog
    developers.openai.com
  3. OpenAIPrimary
    GPT-6.2 Astra model page
    developers.openai.com
  4. OpenAIPrimary
    GPT-6.2 Sol model page
    developers.openai.com
  5. The Decoder
    OpenAI launches GPT-6.2 Astra and Sol with lower API prices and some benchmark gains
    the-decoder.com

How we choose and verify sources

Tags:
  • model selection
  • api pricing
  • context windows
  • coding agents
  • responses api
  • prompt caching
Companies:
  • OpenAI
Models:
  • GPT-6.2 Astra
  • GPT-6.2 Sol
OpenAI launches GPT-6.2 Astra and Sol: prices, specs, caveats · Promptea