GPT-6.2 Astra and Sol: What's New, Pricing, and Benchmarks
OpenAI's newest frontier and everyday models, what changed, and which one to pick.

OpenAI just shipped GPT-6.2 Astra and GPT-6.2 Sol, a paired update to its GPT-6 lineup across ChatGPT Work, Codex, and the API. Astra is the frontier model for the hardest work; Sol is the efficient, versatile model for everyday tasks. Both finish demanding jobs with fewer output tokens, and Astra's API price drops. Here's what actually changed, the pricing, the benchmark claims, and how to choose between them.
What GPT-6.2 Astra and Sol are
OpenAI released both models on November 17, 2026, available the same day in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users, with API access via gpt-6.2-astra and gpt-6.2-sol (OpenAI announcement).
- GPT-6.2 Astra — a frontier model built for your hardest projects. OpenAI says it sets new highs on evals across professional work and scientific research at less than a third of the cost per task of GPT-6 Astra. You'll see the difference in outputs like slideshows, websites, and 3D designs.
- GPT-6.2 Sol — an efficient, versatile model for everyday work. It's tuned for multistep agentic workflows and natural writing, and OpenAI says many of its own employees use Sol as their go-to model.
This continues the cadence OpenAI set earlier in the generation: it kicked off GPT-6 with Astra in September, then upgraded Sol to GPT-6.1 at DevDay. GPT-6.2 brings both tiers forward together.
Pricing: Astra gets cheaper, Sol holds the line
The headline on cost is that Astra's standard input and output prices are 20% lower than GPT-6 Astra, while Sol keeps its GPT-6.1 Sol pricing (9to5Mac).
| Model | API price (per 1M input / output) | vs. predecessor |
|---|---|---|
| GPT-6.2 Astra | $8 / $40 | 20% lower than GPT-6 Astra |
| GPT-6.2 Sol | $2 / $10 | Same as GPT-6.1 Sol |
Beyond token price, OpenAI's pitch is cost per task: because both models reach results with fewer output tokens, you get more done inside the same usage limits. For Astra specifically, OpenAI frames the real-world savings as "less than a third of the cost per task" of GPT-6 Astra on demanding work.
Benchmarks: a solid step, not a reinvention
OpenAI reports steady gains rather than a dramatic leap:
- GPT-6.2 Sol improves 5.3 points over GPT-6.1 Sol on OpenAI's reported evals.
- On Cognition's FrontierCode 2.0 — which measures whether changes to real codebases are good enough for maintainers to merge — OpenAI says that with both models at maximum reasoning effort, Sol came within half a point of Astra at about one-fifth the cost (9to5Mac).
- On the Codex side, OpenAI says Astra is better at making and testing changes in an existing project, and that "early testers reported cleaner code style and more focused tests, with fewer rounds of revision."
As always with vendor-reported numbers, treat first-party benchmarks as directional until independent evaluations land. But the Sol-vs-Astra gap on FrontierCode 2.0 is the number worth watching: if a near-Astra coding score really costs a fifth as much, Sol becomes the default for most agent loops.
Ultrafast comes to more plans
OpenAI expanded Ultrafast, its higher-throughput mode:
- Pro 500 subscribers can use Ultrafast with either model.
- The $200/month Pro 200 plan gets Ultrafast for the first time, starting with GPT-6.2 Sol.
- Developers get Ultrafast on both models through the API at launch.
OpenAI says Ultrafast can generate tokens at up to 4.5 times the standard rate on either model, at the cost of using more of your allowance (9to5Mac). You switch speeds from the model picker.
Safety: tighter safeguards at the frontier
OpenAI says GPT-6.2 Astra and Sol mark another jump in cyber capabilities, and that — like GPT-6 Astra — both meet the Critical threshold for cybersecurity under its Preparedness Framework. OpenAI also says both meet the High threshold for biological and chemical capabilities, and that it's launching the models with tighter safeguards for higher-risk cybersecurity assistance. If you're building autonomous agents on these models, plan for the extra guardrails around sensitive tasks.
GPT-6.2 Astra vs Sol: which one to pick
Think of it as a tiering decision, the same pattern we described for GPT-6 Sol and Luna:
- GPT-6.2 Astra — reach for it on the hardest professional and scientific work, complex Codex changes, and polished visual outputs like slides, sites, and 3D designs.
- GPT-6.2 Sol — the everyday workhorse for agentic workflows, routine coding, and writing. With a near-Astra coding score at roughly one-fifth the cost, it's the sensible default for most multistep agent runs.
A practical pattern: run standard agent loops on Sol, and escalate only the hardest subtasks to Astra. The cost gap makes that kind of model routing worth building. For the cheaper computer-use tier in the same family, see GPT-6.1 Terra.
Availability at a glance
- ChatGPT Work and Codex: both models live today for Plus, Pro, Enterprise, and Edu.
- API:
gpt-6.2-astraandgpt-6.2-sol, with Ultrafast available for both at launch. - Ultrafast access: Pro 500 on both models; Pro 200 starting with Sol; eligible Enterprise and Edu workspaces once an admin enables it.
Put GPT-6.2 to work with your own AI workforce
The real win in GPT-6.2 is cost-efficient agentic work: a near-frontier coding model at a fraction of the price makes model routing a concrete cost lever. Eigent is an open-source multi-agent "Cowork" desktop app that turns models like Sol and Astra into an AI workforce that does the work locally — writing code, running multistep tasks, and automating workflows. If your team ships software, pair the new models with an agent that can review GitHub PRs end to end. Download Eigent and route the cheap tier to the routine steps.
Recent Posts

Eigent Release Notes v1.0.8: A New Home, Global Search & Clearer Conversations
Eigent v1.0.8 introduces a redesigned Home, search across Spaces, a new composer, and a default conversation timeline that makes agent work easier to follow.

What Is NVIDIA Nemotron for Science? Open Models for Research Agents
NVIDIA Nemotron for Science explained: the open model, LitmusBench v0.2 chemistry dataset, and post-training recipes for building AI agents for scientific research.

Claude Fable 5.2 and Mythos 5.2: What's New, Explained
Claude Fable 5.2 and Mythos 5.2 explained: lower pricing at $8/$40 per million tokens, up to 30% faster output, new benchmarks, and who can access each tier.




