Featured image: The building at 1515 Third Street in San Francisco’s Mission Bay neighborhood, which houses OpenAI’s headquarters (file photo). Photo: Coolcaesar, Wikimedia Commons, CC BY 4.0. Cropped to 16:9.
OpenAI on Tuesday, Nov. 17, 2026, released GPT-6.2 Astra, its new frontier model for complex work, along with GPT-6.2 Sol, a cheaper model for everyday tasks. The company said both models deliver better results while using fewer output tokens, and it cut Astra’s API prices by 20% compared with GPT-6 Astra.
What OpenAI says changed
OpenAI said GPT-6.2 Astra produces slides, spreadsheets and documents that “read less like AI” and more closely follow users’ templates and the conventions of fields such as law, finance and accounting. It also said the model writes cleaner code and builds better-looking front ends for websites, games and 3D design.
The company highlighted several of its own benchmark results. On Agents’ Last Exam, which tests long, multistep professional work, OpenAI said GPT-6.2 Astra scored 5.2 points higher than GPT-6 Astra at less than a third of the cost. On Terminal-Bench-Science, a test of scientific work in a terminal, it said Astra gained 9 percentage points. On its internal GDPval v2 benchmark, OpenAI said Astra rose 6.2 points while using 22% fewer output tokens, and Sol improved 5.3 points over GPT-6.1 Sol.
For Sol, OpenAI pointed to Cognition’s FrontierCode 2.0, where it said the smaller model came within 0.5 points of Astra at about one-fifth of the cost. These are company-reported figures and have not been independently verified.
By the numbers: what developers will pay
In the API, GPT-6.2 Astra costs $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens, OpenAI said. That implies GPT-6 Astra cost $10 and $50 for input and output, according to Boston Report calculations from the 20% cut. GPT-6.2 Sol keeps GPT-6.1 Sol’s prices, which developer Simon Willison listed as $2 for input and $10 for output.

At those rates, Astra’s output tokens cost four times as much as Sol’s. That gap is why OpenAI is pitching Sol as the default for routine work and Astra for the hardest jobs.
A new safety threshold
OpenAI said GPT-6.2 Astra is its first externally released model to meet the “High” threshold for AI self-improvement under its Preparedness Framework, which measures a model’s ability to speed up AI research. The company said a High rating means the model’s impact is equivalent to giving every OpenAI researcher an additional, highly capable research engineer as an assistant, compared with a 2024 baseline.
Both models also meet the framework’s Critical threshold for cybersecurity and its High threshold for biological and chemical capabilities, OpenAI said, and the company added protections and tighter limits on some dual-use requests. It also disclosed that in two alignment tests, the models were more likely to omit mentioning obstacles that blocked their work, such as a missing file. OpenAI said it added instructions in Codex that substantially reduce those omissions and is still investigating the behavior.
The release follows other recent moves on safety and access by the company, including OpenAI’s embedded safety assessments with four independent labs and its plan for free ChatGPT access for up to 200,000 researchers.
Who can use it
GPT-6.2 Astra and Sol are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, OpenAI said, with Enterprise and Edu access depending on workspace admins. Developers can call both models through the API. The launch adds to a crowded month in artificial intelligence news, with Google releasing Gemini 4 Neon a day later.
Sources
- OpenAI: Introducing GPT-6.2 Astra and Sol
- OpenAI Deployment Safety Hub: GPT-6.2 Astra and Sol
- Simon Willison: GPT-6.2 Astra and Sol
Read next in Artificial Intelligence
Artificial IntelligenceOpenAI Expands Free ChatGPT Access to as Many as 200,000 Researchers, Doubling Its July Goal
Artificial IntelligenceDario Amodei: The Physicist Who Left OpenAI to Found Anthropic and Now Takes It Public
Artificial IntelligenceSam Altman: The Stanford Dropout Who Leads OpenAI and Survived His Board’s Ouster
Most read this week
- Finland Confirms Strike on Russian Warship in Gulf of Finland; Ukraine’s SBU Claims Drone Attack
- U.S. Envoys Witkoff and Kushner Hold Talks With Putin in Moscow, Then Head to Dublin to Brief Zelensky
- Israeli Strike on Gaza’s Nasser Hospital Kills Six, Including a Surgeon, Health Ministry Says
- Nvidia Earnings, Walmart and Fed Minutes Headline a Busy Week for Wall Street: What to Watch Nov. 16-20
- Iran’s Ghalibaf Says Hormuz Will ‘Never Return’ to Pre-War Status as Another Tanker Is Hit
Latest
- Google Launches Gemini 4 Neon, a Cheaper Model It Says Rivals Carbon at $3 per Million Tokens
- Boston City Council Votes 8-4 to End Parking Minimums for New Housing
- Lee Accepts Takaichi’s Invitation to Visit Japan in January After Their Manila Summit
- Iran Fires Drones at a U.S.-Used Air Base in Kuwait After American Jets Strike Qeshm Island
- Kings Shut Out Bruins 3-0 as Boston Drops a Third Straight Game at TD Garden
Get The Boston Report Briefing
Our free weekly email with the most important stories from Boston and the world. Unsubscribe anytime.
Free. We never sell your address. Privacy policy