GPT-6.2 Astra and Sol: what's new, pricing, how to use
2026-11-21 · 4 min read
GPT-6.2 Astra and GPT-6.2 Sol are now available. Astra is OpenAI's new flagship, Sol is the cheaper, faster sibling, and both are open to paid ChatGPT users, Codex and the API.
@OpenAI said they are in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and that Pro 500 subscribers can use an "Ultrafast" mode with either model. Pro 200 subscribers get Ultrafast too, starting with Sol. @OpenAIDevs confirmed both models are in Codex and the API.
Our earlier piece on the GPT-6.2 Sol rumor covered users guessing they were already being routed to it. Now it is official.
What OpenAI says changed
The main claim from @OpenAIDevs is that Astra delivers "cleaner code, better frontends and 3D models, and more efficient tool use." Both models are meant to get better results on hard tasks with fewer output tokens, which matters because output tokens are what you pay for.
A second post from @OpenAIDevs cites Terminal-Bench Science 0.1, a test of scientific work done in a terminal. It says Astra scores 9 percentage points above GPT-6 Astra at under a third of the cost. That is OpenAI's own number, so treat it as a vendor claim until independent runs show up.
@sama said the models are very token-efficient and singled out Sol with Ultrafast as the one he expects people to enjoy.
The first outside reaction is mixed
Not everyone is impressed. @danshipper of Every said his team had tested both for a few weeks and called it "a disappointing release compared to Fable 5.2." He found Astra 6.2 a step back from Astra 6 on writing.
That is one team's experience, and writing quality is a narrow slice. It is still a useful warning: a model that wins on code and tool use can lose on prose. If you want the comparison point, see our post on Claude Fable 5.2.
What it costs
Pricing figures below come from third-party trackers, not from a post I could open on X, and aggregators disagree in places. Confirm on OpenAI's pricing page before you budget.
| Model | Input per 1M tokens | Output per 1M tokens | Where it fits |
|---|---|---|---|
| GPT-6.2 Astra | $8 | $40 | Hard coding and reasoning jobs |
| GPT-6.2 Sol | $2 | $10 | High-volume, cost-sensitive work |
These rates are reported for prompts up to about 272K input tokens. Ultrafast has its own pricing that I could not confirm, so check it before turning it on for a batch job.
How to try it
- In ChatGPT, open the model picker on a Plus or Pro plan and look for Astra and Sol. Rollouts can be staggered, so if you do not see them yet, check again later.
- In Codex, pick the model from the model setting and run the same task you gave your current default.
- In the API, call
gpt-6.2-astraorgpt-6.2-sol.
A fair test costs little. Take three real tasks from your week, say a messy email thread to summarize, a spreadsheet formula to fix, and a customer reply to draft. Run each on Sol first, then Astra, and weigh quality against price. My guess is Sol covers most small-business work, but that is a guess, not a benchmark result.
Keep a copy of the old model's output too. Side by side, you can spot a regression like the writing slide Shipper described instead of trusting a leaderboard.
What this means for a small business
Model launches come every few weeks now, and which one wins depends on the job. Don't chase each release. Keep a small set of real tasks you can rerun in ten minutes whenever a new model lands.
Reports around the launch also mention tighter safeguards on higher-risk cybersecurity requests. We covered the earlier debate in how Astra shipped with a critical cyber rating. If your staff use these tools, decide now what data can go into them.
If you are not sure which of your workflows are worth automating with a model like this, New Face Design offers a free process audit. We will look at how your team works today and tell you where AI would actually save time.