
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, adding two options for software development and tool-using agents. Astra targets demanding reasoning and coding; Sol targets everyday coding and agent workflows. Both support Responses and Chat Completions, but tool calling requires the Responses API. The official changelog makes that endpoint distinction explicit.
What changes in Standard pricing
The clearest commercial change is Astra's pricing:
- GPT-6.2 Astra: $8 per million input tokens and $40 per million output tokens—20% below GPT-6 Astra's $10/$50 Standard rates.
- GPT-6.2 Sol: $2 input and $10 output per million tokens, unchanged from GPT-6.1 Sol.
These Standard rates apply to requests with up to 272,000 input tokens. Longer requests cost more, with separate charges for tools and other processing tiers. The official pricing page is the reference for the request size and processing mode being used.
Context and agent capabilities
Both models accept text and images, produce text, and offer a 1.05-million-token context window. Their documented agent capabilities include web search, computer use, and delegation through the Responses API's Multi-agent beta. These are supported capabilities, not a claim that every tool was newly introduced with this release.
The Astra documentation and Sol documentation also state that the API rejects the none reasoning-effort setting. Existing integrations should check model-specific settings before switching IDs.
Why it matters for builders
For builders, the useful distinction is demanding work versus routine execution. Astra's lower token rates make it worth reevaluating for difficult coding tasks, while Sol offers a lower-priced option for repeated agent workflows.
For teams like SunMarc App Labs, the practical test is real app fixes and browser-based QA: compare successful completion, retries, and total cost—not just token prices. A cheaper token rate does not establish a cheaper completed task if extra retries or manual corrections are needed.
We have not tested these releases. This is a proposed evaluation, not a measured performance or cost result. For earlier context, see our GPT-6.1 Sol coverage.