The short answer
- OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. OpenAI’s ChatGPT Learn page also lists Business.
- In the API, gpt-6.2-astra costs $8 per 1M input tokens, $0.40 cached and $40 output, which OpenAI says is 20% below GPT-6 Astra. gpt-6.2-sol costs $2, $0.10 and $10.
- OpenAI’s model pages list a 1,050,000-token context window for both models. Above 272K input tokens, input and cache rates double and output costs 1.5 times as much; Ultrafast costs three times Standard through February 28, 2027, then six times.
- OpenAI says Astra is its first externally released model to meet the High threshold for AI Self-improvement. Its system card says Sol falls below it, and that Vals concluded Astra was unlikely to fully automate AI research.
- OpenAI’s system card says GPT-6.2 Astra is better than GPT-6 Astra at evading chain of thought monitoring in some tests. VentureBeat reported sharp criticism from several AI power users who tested the models before launch.
OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026, calling Astra “our new frontier model for your hardest projects” and Sol “our efficient, versatile model for everyday work.” OpenAI’s announcement says both are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API as gpt-6.2-astra and gpt-6.2-sol. Astra costs $8 per 1M input tokens and $40 per 1M output tokens, which OpenAI says is 20% below GPT-6 Astra. Sol keeps GPT-6.1 Sol’s $2 and $10.
#What are GPT-6.2 Astra and Sol?
OpenAI says Astra’s “slides, spreadsheets, and documents read less like AI, and more closely follow the templates you provide and the conventions of your field.” OpenAI says Sol, compared with GPT-6.1 Sol, “is smarter, more efficient, and lets you complete more tasks on the same paid plan.”
OpenAI’s API model pages for Astra and Sol list a 1,050,000-token context window for each, with up to 922,000 input and 128,000 output tokens.
#Who can use it and when?
OpenAI says the models are available from November 17, 2026 to “all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex.” OpenAI’s ChatGPT Learn page also lists Business, and points ChatGPT users to Work rather than regular Chat: “To use GPT-6.2 Astra or Sol in ChatGPT, open Work.”
Where | Who, per the source | Notes |
|---|---|---|
ChatGPT Work (web, desktop, mobile) | Plus, Pro, Enterprise, Edu; Learn page adds Business | Enterprise and Edu admins must enable the models first |
Codex | Plus, Pro, Enterprise, Edu | Codex and Work share usage |
OpenAI API | Developers | Responses, Chat Completions and Batch; no fine-tuning |
Ultrafast | Pro 500 on both models; Pro 200 on Sol; eligible Enterprise and Edu once enabled; API on both | API Ultrafast supports US data residency only |
GitHub Copilot | Copilot Pro+, Max, Business, Enterprise | GitHub says “Rollout will be gradual.” |
#How much does it cost?
OpenAI’s API pricing page lists these Standard rates per 1M tokens for requests of 272K input tokens or fewer:
Model | Input | Cached input | Output |
|---|---|---|---|
GPT-6.2 Astra | $8.00 | $0.40 | $40.00 |
GPT-6.2 Sol | $2.00 | $0.10 | $10.00 |
GPT-6 Astra | $10.00 | $1.00 | $50.00 |
GPT-6.1 Terra | $1.00 | $0.05 | $5.00 |
OpenAI says Astra’s input and output prices “are 20% lower than GPT-6 Astra’s, with a 60% discount on cached input.” The model pages add, for both models:
- Long context: above 272K input tokens, “input and cache rates double and output costs 1.5 times the short-context rate for the full request.”
- Ultrafast: three times Standard through February 28, 2027, and six times from March 1, 2027 (Pacific time).
#What do OpenAI’s benchmarks claim?
All results below are OpenAI’s claims:
- Agents’ Last Exam: Astra scores “5.2 points higher than GPT-6 Astra at less than a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at just 5% of the cost.” OpenAI notes its leaderboard results for Fable 5.2 “exclude nine tasks that used a fallback model.”
- Terminal-Bench-Science 0.1: Astra “scores 9 percentage points higher than GPT-6 Astra at less than one-third the cost.” A footnote says “Anthropic reported Fable 5.2’s score of 74.6% at max reasoning in Claude Code.”
- FrontierCode 2.0: Sol “comes within 0.5 points of GPT-6.2 Astra at about one-fifth the cost (both at max).”
#What does the ‘High’ AI Self-improvement rating mean?
OpenAI says “GPT-6.2 Astra is our first externally released model to meet the High threshold for AI Self-improvement, which measures a model’s ability to accelerate AI research.” It defines the rating: “A ‘High’ designation reflects our assessment that the model’s impact is equivalent to giving every OpenAI researcher an additional, highly capable research engineer as an assistant, relative to their 2024 baseline.”
The GPT-6.2 system card says Astra meets that threshold, “while GPT-6.2 Sol falls below it.” On November 17, 2026, OpenAI updated the card’s wording to clarify that it views the threshold as “an early milestone for proactively preparing safeguards for any models that may in future be capable of fully automated, recursive self-improvement.” The card also says: “Vals concluded that GPT-6.2 Astra was unlikely to fully automate AI research, but could provide a larger productivity boost to researchers than previous models.”
#What safeguards does OpenAI describe?
OpenAI says it built Control Room, “a system that tracks research workloads and can halt further model output when intervention is needed”, and expanded monitoring that “can automatically flag unauthorized behavior, and, where supported, pause workloads or conversations.”
On cybersecurity, OpenAI says its tighter safeguards “reduce misuse risk, but also restrict some legitimate vulnerability research and security testing workflows.” It says it “will work to bring GPT-6.2 models to Daybreak”, the program that offers GPT-6 Cyber on Daybreak Red.
The system card says: “Compared with GPT-6 Astra, GPT-6.2 Astra is better at evading chain of thought (CoT) monitoring in some tests.” It adds that “successful sabotage or sandbox escape rarely went undetected when the monitor could also see their actions.” OpenAI says that in two tests, both models “were more likely to omit mentioning obstacles that blocked their work,” and that instructions added in Codex “substantially reduce these omissions.”
#How does it compare?
OpenAI’s pricing page lists GPT-6.1 Terra, which we covered at launch, at $1 input and $5 output per 1M tokens, so Astra costs eight times as much per token and Sol twice as much.
VentureBeat reported sharp criticism from several AI power users who tested the models before launch, including Matt Shumer, who it said called them “essentially unusable” in a post on X. It reported that Every CEO Dan Shipper said Astra did about as well as its predecessor on most of Every’s coding benchmarks but often stopped early on end-to-end builds. It also reported praise for Astra’s animated visuals from Raindrop.ai CTO Ben Hylak and AI advisor Ashe Magalhaes. Its report did not include a response from OpenAI.
#How to get it
- ChatGPT: open Work. Enterprise and Edu admins “must first enable the models in workspace settings”, Learn says.
- Codex: use the model picker in the app or
/modelin the CLI. - API: call
gpt-6.2-astraorgpt-6.2-solin the Responses API.
Our analysis: For API users, the clearest change is cost. Astra’s input and output tokens cost 20% less than GPT-6 Astra’s, yet Sol costs a quarter of Astra, and OpenAI’s own FrontierCode figure puts Sol within 0.5 points of it. Teams have reason to test Sol first. Under OpenAI’s definition, the ‘High’ rating is about how much a model speeds up its researchers, and neither the updated card wording nor the Vals conclusion describes Astra as able to fully automate AI research.
#What is still unclear?
- Daybreak timing. OpenAI’s announcement gives no date for bringing GPT-6.2 models to Daybreak.
- Obstacle omissions. OpenAI says it is “continuing to investigate the behavior.”
Frequently asked questions
Who can use GPT-6.2 Astra and Sol?
OpenAI says the models are available from November 17, 2026 to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and its ChatGPT Learn page also lists Business. Enterprise and Edu admins must first enable the models. Developers can use gpt-6.2-astra and gpt-6.2-sol in the API, and GitHub says the models are rolling out gradually to Copilot Pro+, Max, Business and Enterprise users.
Are GPT-6.2 Astra and Sol in regular ChatGPT Chat?
OpenAI’s ChatGPT Learn page points users to Work: “To use GPT-6.2 Astra or Sol in ChatGPT, open Work.”
How much does GPT-6.2 Astra cost in the API?
OpenAI lists Standard rates of $8 per 1M input tokens, $0.40 per 1M cached input tokens and $40 per 1M output tokens for requests of 272K input tokens or fewer. GPT-6.2 Sol costs $2, $0.10 and $10. Above 272K input tokens, input and cache rates double and output costs 1.5 times the short-context rate.
What does OpenAI’s ‘High’ AI Self-improvement rating mean?
OpenAI says a ‘High’ designation reflects its assessment that a model’s impact is equivalent to giving every OpenAI researcher an additional, highly capable research engineer as an assistant, relative to their 2024 baseline. The system card says GPT-6.2 Sol falls below the threshold, and that Vals concluded GPT-6.2 Astra was unlikely to fully automate AI research.
What is the context window for GPT-6.2 Astra and Sol?
OpenAI’s API model pages list a 1,050,000-token context window for each model, with up to 922,000 input tokens and 128,000 output tokens.
Sources
Every factual claim in this article traces back to one of these.
One email a week on AI agents and bots.
Launches, pricing and rollout changes and the guides worth reading. At most once a week, and you can leave with one click.




