On November 17, 2026, OpenAI rolled out GPT-6.2 Astra and an updated GPT-6.2 Sol inside ChatGPT Work and Codex, available to Plus, Pro, Enterprise, and Edu users on those two surfaces. If you only use regular ChatGPT and don’t see them, that’s expected: Astra and Sol are rolling out through ChatGPT Work and Codex specifically, not the standard chat interface, and availability within those surfaces still depends on your plan, workspace permissions, and where OpenAI’s rollout currently stands. If you’re on one of those plans and using ChatGPT Work or Codex, the short answer to “does this change how I use it?” is mostly yes, in a good way.
OpenAI says both models finish tasks with fewer output tokens. Fewer tokens can help you stretch your plan’s usage further, though the exact effect depends on your plan, the model, and whether you’re using Ultrafast mode.
What Happened With GPT-6.2 Astra and Sol
OpenAI announced the pair on its official GPT-6.2 announcement page, and The Verge and 9to5Mac both reported availability the same day.
Here’s the plain-English split between the two:
- GPT-6.2 Astra: OpenAI’s top-tier “frontier” model for your hardest work. OpenAI says it follows real templates more closely, matching layout, colors, fonts, and tone in slides, spreadsheets, and documents. It’s also pitched for coding, website frontends, games, and 3D design.
- GPT-6.2 Sol: OpenAI’s everyday model, positioned as efficient for writing, coding, and multistep tasks. OpenAI also says many of its own employees default to it.
- Ultrafast mode: A new optional speed setting that OpenAI says delivers noticeably faster responses. The catch? It burns through your allowance quicker. Pro 200 gets Ultrafast with GPT-6.2 Sol, while Pro 500 gets it with both Astra and Sol.
- Enterprise and Edu: Access may roll out gradually across eligible workspaces, and eligible workspaces can also get Ultrafast once an admin turns it on. If you don’t see the new models yet, check with your workspace admin.
Think of Sol as your daily driver and Astra as the moving truck you rent for the big job. You can switch between them from ChatGPT Work’s model picker, when your account and workspace have access; Codex offers its own model-selection option, which may not work exactly the same way. Either way, there’s nothing to download.
How It Stacks Up
For developers and businesses paying per use through the API, OpenAI’s launch announcement lists GPT-6.2 Astra at $8 per million input tokens, $0.40 per million cached input tokens, and $40 per million output tokens for standard-context requests. GPT-6.2 Sol’s announced rate is $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. Other tiers, such as long-context or faster-response processing, can carry different rates.
A token is a chunk of text, roughly a short word, and that’s how API usage is billed. These figures are OpenAI’s own announced launch-day rates for standard requests; check OpenAI’s pricing page directly for the full tier breakdown, including any long-context or faster-processing rates, before budgeting.
| Model | Best for | API pricing (per 1M tokens) | Ultrafast access |
|---|---|---|---|
| GPT-6.2 Astra | Complex documents, data analysis, big coding and design projects | $8 input / $0.40 cached input / $40 output | Pro 500 only |
| GPT-6.2 Sol | Everyday writing, planning, light coding | $2 input / $0.10 cached input / $10 output | Pro 200 (Sol) and Pro 500 (both models) |
See OpenAI’s pricing page for the complete tier breakdown, since standard, long-context, and faster-processing rates can differ and change.
So what about Claude? OpenAI reports that GPT-6.2 Astra performs strongly against Anthropic’s Claude Fable 5.2 on several internal benchmarks. Treat those results as vendor-reported, since OpenAI ran and published them itself. The same goes for OpenAI’s other benchmark claims, including the science and data-analysis ones. Comparisons against Google’s Gemini models will have to wait for independent testers.
What you can compare today is price: at OpenAI’s announced standard-tier rate, Astra runs $8 input and $40 output per million tokens via the API, while Sol runs $2 input and $10 output. For ChatGPT subscribers, Ultrafast access depends on both plan and model.
Pro 200 includes Ultrafast for Sol, while Pro 500 includes it for both Astra and Sol. It’s not a perk of Plus or Enterprise/Edu by default, though eligible workspaces can get it once an admin enables it.
The Reaction
Reaction on X is still forming. Early posts mostly repeat the pricing and availability details, and a few people say they plan to try the models. As of this writing, Hacker News doesn’t have a thread on the release yet.
Our Take
If you’re on ChatGPT Work or using Codex and mostly write emails, summarize notes, or plan trips, stick with Sol. OpenAI pitches it as the efficient default, and that should help stretch your usage further if the efficiency claims hold up.
Reach for Astra when the output has to look polished. A client deck built on your company template or a messy spreadsheet analysis are good examples.
Freelancers and small business owners on ChatGPT Work have the most reason to care, since Astra’s template-matching pitch targets the tedious cleanup work. Codex users get both models for coding projects too.
I’d skip upgrading to Pro 500 just for Ultrafast on Astra unless waiting on responses is a daily frustration. If you mainly work in Sol, Pro 200 already includes Ultrafast access.
Faster answers that drain your allowance sooner is a trade, and it’s only worth it when speed matters more than volume. Developers comparing costs should note Astra’s new announced standard-tier rate of $8 input, $0.40 cached input, and $40 output per million tokens, versus Sol’s unchanged $2/$0.10/$10 — check OpenAI’s pricing page if your workload uses long-context or faster-processing requests, which carry different rates.
Wrapping Up
GPT-6.2 Astra and Sol arrived in ChatGPT Work and Codex on November 17, 2026, available to Plus, Pro, Enterprise, and Edu users on those surfaces. Default to Sol for everyday tasks and save Astra for high-stakes deliverables.
Check your Pro tier before hunting for Ultrafast: Pro 200 covers Sol, and Pro 500 covers both models. Eligible Enterprise and Edu workspaces can also get it once an admin flips it on.
My honest take: the efficiency angle sounds promising for anyone who keeps bumping into usage limits. Still, every performance claim here comes from OpenAI. Hold off on any big plan changes until independent testers weigh in.