OpenAI launches GPT-6.2 Astra and Sol — cheaper, faster AI models for work and daily use
Two new models from OpenAI promise better results for a fraction of the cost, making expensive AI tools more affordable for businesses and everyday users.

What you could do with it
- A lawyer could review contract provisions across source documents and produce deliverables meeting every requirement, using Astra for Law at roughly one-sixth the cost of rival models 1.
- An office worker could create professional slide decks that match exact layouts, colors, and typography from an existing file or screenshot, then extend the template's distinctive writing style to new slides 1.
- A scientist could carry out multi-step research tasks inside a command-line terminal, with GPT-6.2 Astra scoring 9 percentage points higher than GPT-6 Astra for less than a third the price 1.
From the official announcementOffice documents
What each model does
GPT-6.2 Astra is aimed at hard professional work. It writes cleaner code, builds better-looking frontends for websites and games, and creates slides, spreadsheets and documents that follow the style of your industry more closely 1.
GPT-6.2 Sol is built for day-to-day tasks. According to OpenAI, numerous workers have adopted it as their go-to model for both technical tasks and everyday assignments, praising its speed, clarity of expression, and collaborative usability 1.
From the official announcementOverview
From the official announcementWebsites
From the official announcement3D worlds and games
Where Astra leads
In OpenAI’s own testing, Astra scored 5.2 points higher than the older GPT-6 Astra on Agents’ Last Exam, a benchmark that measures how well models handle long multi-step professional workflows, at less than a third of the cost 1. It also beat Claude Fable 5.2 by 5.5 points on the same exam for just 5 percent of the cost . On the Terminal-Bench-Science benchmark, which tests a model’s ability to do scientific work inside a command-line terminal, Astra scored 9 percentage points higher than GPT-6 Astra for less than a third of the price . It also ranked first on the Vals Science Index, an evaluation that requires models to interpret actual biological data and conduct simulated experiments .
For legal work, Astra more than doubled the pass rate of Claude Fable 5.2 on the Harvey LAB-AA benchmark, where the model must find relevant facts across source documents and produce a deliverable meeting every requirement, at roughly one-sixth the cost 1. Astra for Law, a configuration tuned for lawyers, scored even higher . On the Artificial Analysis AA-AnalystAgent test, which uses real spreadsheets and reports — such as California Medicaid expenditure data to calculate spending on medical transport — Astra answered 62.5 percent of questions correctly across five independent attempts .
On the internal GDPval v2 benchmark, which covers long knowledge-work assignments from consulting, finance and design, Astra scored 6.2 percentage points above GPT-6 Astra while using 22 percent fewer output tokens 1.
Astra also impressed early testers on slide design. When provided with an original file—or even a screenshot of a slide deck—it replicates the exact layout, colors, and typography including precise font sizing, and extends the template's distinctive writing style to newly created slides 1. Documents were described as significantly clearer than before, making it easier to pick out key points from briefing materials .
How Sol compares
Sol stays close to Astra’s level on coding tasks without the premium price tag. On the FrontierCode 2.0 benchmark, which checks whether changes to real codebases meet maintainers’ standards, Sol came within 0.5 points of Astra at about one-fifth the cost 1. On the LifeVal evaluation, which measures performance on personal everyday tasks like planning meals or researching schools, Sol outscored GPT-6 Astra at roughly one-seventh the cost . Sol also improved 5.3 points over GPT-6.1 Sol on GDPval v2 .
Price changes
In the API, Astra’s input and output token prices are 20 percent lower than GPT-6 Astra. Input costs $8 per million tokens, cached input (repeated content) drops to $0.40 per million, and output is $40 per million 1. Sol keeps the same token prices as GPT-6.1 Sol .
This story reports on a trend for information only. It is not advice or encouragement to do any of it.
How good is it
1 According to OpenAI , GPT-6.2 Astra finished the Agents' Last Exam—a challenge testing long multi-step professional workflows—5.2 points ahead of GPT-6 Astra for under a third of the cost, and outperformed Claude Fable 5.2 by 5.5 points at only 5 percent of its price. On the Harvey LAB-AA legal benchmark, Astra more than doubled Claude Fable 5.2's pass rate at roughly one-sixth the cost . 2 On terminal-based science work, Astra came in 9 percentage points above GPT-6 Astra at less than one-third the price, establishing new records on both Terminal-Bench-Science and the Vals Science Index . For everyday tasks measured by LifeVal, GPT-6.2 Sol achieved higher scores than GPT-6 Astra at about one-seventh the cost .
From the official announcementMore from the announcement
Key facts
- GPT-6.2 Astra is 20% cheaper on API input/output than GPT-6 Astra
- Astra scored 5.2 points higher than GPT-6 Astra on Agents' Last Exam at less than a third the cost
- GPT-6.2 Sol costs about one-seventh of GPT-6 Astra while outscoring it on everyday task benchmark LifeVal
- Both models use fewer output tokens to achieve better results
- Astra set new highs on Terminal-Bench-Science 0.1 and Vals Science Index benchmarks
Sources
If this story concerns you or your company, or uses something that belongs to you, ask for a correction or for it to be taken down. How requests work
Keep reading
Claude Fable 5.2 scores 65 on Artificial Analysis — highest intelligence rating yet
Anthropic's latest Claude version tops a widely tracked AI leaderboard, but it comes with steep prices and long wait times.
How it worksOpenAI Releases 372 AI-Generated Math Proofs on GitHubHerald of AI
OpenAI Releases 372 AI-Generated Math Proofs on GitHub
The company is bypassing traditional journals, betting that formal verification can handle the flood of AI-produced mathematics.
OpenAI’s ChatGPT Wallet lets AI agents shop without seeing your card number
It marks the start of a new way to buy things online — your AI helper picks and pays, but you still approve every transaction and your real card stays hidden.