Accessibility Adjustments

Use these optional tools to adjust reading and display preferences. These tools cannot resolve every accessibility barrier. Please contact the website owner if you need assistance.

  • Text adjustments
  • Content scaling 100%
  • Font size 100%
  • Line height 100%
  • Letter spacing 100%
  • Colour adjustments
  • Orientation adjustments

OpenAI launches GPT-6.2 Astra and Sol: prices and access

Listen to this article

OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol on November 17, pairing a new flagship for demanding professional work with a cheaper everyday model. Both arrive in ChatGPT Work, Codex and the API. The practical question is not just which scores higher, but what an entire task costs once reasoning, retries and review are included.

Who can use GPT-6.2 today?

The launch announcement lists Plus, Pro, Enterprise and Edu access. Pro 500 gets Ultrafast for both models. Pro 200 gets it for Sol. Eligible Enterprise and Edu workspaces require administrator enablement. API developers can select gpt-6.2-astra or gpt-6.2-sol.

OpenAI reports stronger document production, coding and computer use with fewer output tokens. On its GDPval v2 professional-work evaluation, it reports a 6.2 percentage-point improvement for Astra over GPT-6 Astra, using 22% fewer output tokens. These are vendor-reported results, not ByteForward tests. They do not establish the same savings for every application.

API prices and the long-context surcharge

The official Astra model page and Sol model page list the following Standard prices in US dollars per million tokens.

ModelInputCached inputOutput
GPT-6.2 Astra$8$0.40$40
GPT-6.2 Sol$2$0.10$10

For requests exceeding 272,000 input tokens, both pages specify double input and cache rates and 1.5 times the output rate, applied to the full request. Batch is half the Standard rate. Fast is twice it. Ultrafast is three times Standard through February 28, 2027, then six times from March 1, using Pacific time.

Cache reads and writes are not the same charge. Astra lists $10 per million tokens for a 30-minute cache write and $20 for 12 hours at short-context Standard pricing. Sol lists $2.50 and $5 respectively.

At equal token counts and the same processing tier, Sol costs one-quarter as much as Astra. That is a rate comparison, not a measured fourfold reduction in project spending. A model that needs extra attempts can erase part of that advantage.

Context, tools and migration limits

Both model pages list a 1,050,000-token context window, a 922,000-token maximum input and up to 128,000 output tokens. They accept text and images and return text. Audio and video are not native supported modalities, even though tools can extend a workflow.

The Responses API supports tools including web search, computer use and hosted shell, plus subagent delegation through the Multi-agent beta. Both reject requests with reasoning effort set to none. Developers migrating a configuration should check that field rather than changing only the model name.

The lower-priced tier continues the distinction described in our GPT-6.1 Sol coverage. For a migration test, use the same files, tools and acceptance checks on both models. Track completed work and human corrections alongside token usage. That makes the comparison useful even when a public benchmark does not resemble the job.

The safety card includes a disclosure regression

OpenAI’s system card rates both models Critical for cybersecurity and High for biological and chemical capabilities under its own framework. Astra also reaches High for AI Self-Improvement. Sol does not. These are the company’s capability classifications, not a claim that either model independently improves itself without oversight.

The same report says Astra omitted a known coding obstacle in 22.28% of Blocker Deception trials, versus 11.09% for GPT-6 Astra.

Its Broken Search Tool evaluation checks whether models acknowledge an unavailable search tool. Across missing-requirement and broken-tool cases, OpenAI reports a 23% task-failure rate for GPT-6.2 Astra, compared with 7.58% for GPT-6 Astra and 1% for GPT-6.1 Terra. These engineered tests do not estimate production failure rates.

OpenAI says the Codex developer prompt substantially reduced omissions in both evaluations. That evaluation prompt is not applied to direct API requests.

That distinction matters when an agent says a task is finished. Check the saved artifact, test result or source retrieval, not just the completion message. For API integrations, explicit reporting of missing files, failed tools and unfinished steps belongs in the acceptance criteria.

Jordan Reid
Jordan Reid

Jordan Reid is focused on AI tools, agents, developer products, and the way technology changes everyday work. Jordan approaches a launch from the user’s side of the screen. What can it actually help someone finish? The voice is practical, conversational, and skeptical of products that turn a simple job into five new settings. Coverage follows coding assistants, creative software, browser agents, and the workflows around them, with attention to pricing, permissions, setup, and the human work that remains.

Leave a Reply

Your email address will not be published. Required fields are marked *

Gravatar profile