Overview
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026 for ChatGPT Work, Codex, and the API. Start with Sol for well-scoped work you repeat. Choose Astra when a task spans several systems or when a weak first pass would cause substantial rework. In ChatGPT, open Work to find the new models as they reach your account. Ordinary Chat has its own model picker and availability. Release announcement, Work and Codex access
| Your task | First model to try | What to check before switching |
|---|---|---|
| Fix a bug across services, then verify the change | Astra | Check whether it finds the cause, covers the affected paths, and runs the relevant tests |
| Build a slide deck or workbook from an existing template | Astra | Check template fidelity and trace figures back to source files |
| Update a focused feature, add tests, or prepare a project brief | Sol | Check that it satisfies the brief and leaves a diff or draft you can review quickly |
What Changed
Astra is cheaper per API token than GPT-6 Astra. Standard input and output rates fell 20%, and cached input fell 60%. OpenAI also reported a 6.2 percentage-point improvement on GDPval v2, a benchmark built from professional assignments. The strongest reason to evaluate Astra is a workflow where quality matters enough to offset a higher rate than Sol. OpenAI release results, API pricing
Sol improves at the same Standard API rates as GPT-6.1 Sol. OpenAI reported a 5.3 percentage-point improvement over 6.1 Sol on GDPval v2. That makes 6.2 Sol a candidate for recurring tasks that already fit the Sol cost tier. Try it on the work you actually repeat and track how many runs need correction. The benchmark figures are OpenAI's results. OpenAI release results, API pricing
Pro $200 gains Sol Ultrafast in Codex. Pro $500 supports Ultrafast for both GPT-6.2 models. On those plans, the eligible 6.2 Ultrafast usage draws down the included Work and Codex allowance at six times the selected model's Standard rate. Reserve that setting for work where a faster response is worth the extra allowance. OpenAI Work and Codex usage guide
Compared With Previous Versions
Prices below are in USD per million tokens for Standard API requests at short-context rates. The input, cached-input, and output rates for 6.2 Astra are four times those for 6.2 Sol. The token counts and retries determine the cost of a completed task. OpenAI API pricing
| Model | API model ID | Input | Cached input | Output |
|---|---|---|---|---|
| GPT-6.2 Astra | gpt-6.2-astra |
$8.00 | $0.40 | $40.00 |
| GPT-6 Astra | gpt-6-astra |
$10.00 | $1.00 | $50.00 |
| GPT-6.2 Sol | gpt-6.2-sol |
$2.00 | $0.10 | $10.00 |
| GPT-6.1 Sol | gpt-6.1-sol |
$2.00 | $0.10 | $10.00 |
For example, 100,000 uncached input tokens and 10,000 billed output tokens cost $1.20 with 6.2 Astra or $0.30 with 6.2 Sol at those rates. The estimate covers those tokens only. Requests above 272,000 input tokens use higher rates for the full request. ChatGPT sign-in in Codex draws from the applicable subscription allowance or credits, while API-key use follows API billing. Astra API specifications, Sol API specifications, Work and Codex billing
Availability and Access
- ChatGPT Work: Open the model picker in Work on a supported client. Availability depends on plan, workspace settings, and rollout progress. Managed workspaces may require an administrator to enable access. OpenAI Work and Codex guidance
- Codex: Choose an available model in the desktop app or enter
/modelin an interactive CLI session. OpenAI lists a 272,000-token default context for Astra and 500,000 for Sol in Codex, while both API listings specify a 1,050,000-token context window. Check the client default before planning a large repository or document workflow. Work and Codex model guide, Astra API specifications, Sol API specifications - OpenAI API: Both models accept text and images and return text. Responses and Chat Completions support these model IDs. Use Responses for tool calls. If an old request sets reasoning effort to
noneorminimal, move tolowand check the endpoint's supported parameters before switching the default. OpenAI migration guide, model specifications
Who Should Upgrade or Wait
If you already run GPT-6.1 Sol on recurring tasks, try GPT-6.2 Sol on a small set of representative jobs at the same API rate. Measure acceptance rate, retries, and time to a usable result before moving traffic.
If you chose GPT-6 Astra for a difficult workflow, try its 6.2 successor alongside Sol. Give both the same inputs and acceptance checks, such as passing tests or a correctly populated template. Compare the cost and time of accepted deliverables.
If your workspace lacks access or your integration uses tool calls through Chat Completions, check workspace rollout or complete the API migration first. Keep the evaluated model in service until the new path meets your acceptance checks. Work and Codex access, API migration guide




