OpenAI Drops GPT-6.2 Astra and Sol: 20% Cheaper API, Smarter Code

OpenAI just upgraded ChatGPT and Codex with GPT-6.2 Astra and Sol. You get better reasoning, faster generation, and a 20% API price cut.

Models · Source: 9to5Mac

What happened

OpenAI just released GPT-6.2 Astra and Sol. Both models are available today across ChatGPT Work, Codex, and the API. Users on Plus, Pro, Enterprise, and Edu plans can access them immediately. Astra remains the heavy hitter for your most demanding projects. It can now take a screenshot of a slide and perfectly match the layout, colors, and font size. Early testers report it writes cleaner code with more focused tests and fewer rounds of revision.

Sol is the lightweight model for everyday coding and writing. It punches way above its weight class. On Cognition's FrontierCode 2.0 benchmark, which measures if code changes are good enough to merge, Sol proved its value. When both models run at maximum reasoning effort, Sol came within half a point of Astra. The kicker is that Sol achieves this at about one-fifth the cost of Astra.

Both models are now more efficient. OpenAI states they can finish demanding tasks using fewer output tokens. This means you can get more work done before hitting your usage limits. OpenAI also rolled out a new Ultrafast mode. It generates tokens up to 4.5 times the standard rate, though it consumes more of your allowance.

Key facts

Why it matters

This release changes the math for anyone building AI features. Astra just got a massive 20 percent price cut on standard API usage. Input tokens dropped from $10 to $8 per million. Output tokens dropped from $50 to $40 per million. Cached inputs saw an even steeper 60 percent reduction, falling from $1 to $0.40 per million. You can now run heavy reasoning tasks and agentic workflows for significantly less money. On the Agents' Last Exam benchmark, Astra scored 5.2 points above the previous GPT-6 Astra at less than one-third the cost per task.

The real story is the narrowing gap between cheap and expensive models. Sol keeps the older GPT-6.1 pricing but delivers near-flagship coding performance. Founders can route the vast majority of their application logic through Sol to save money. You only need to reserve Astra for the absolute hardest edge cases. Furthermore, both models now meet OpenAI's Critical cybersecurity threshold. They include tighter safeguards for higher-risk cybersecurity assistance, making them safer for enterprise deployments.

For builders

Route everyday coding tasks to Sol

Sol costs just $2 per million input tokens and $10 per million output tokens. Cached inputs are incredibly cheap at $0.10 per million. It nearly matches Astra on real-world codebase merges at a fraction of the price. Builders who switch their default routing to Sol will drastically cut their API bills without sacrificing much quality.

Leverage Astra for visual UI templating

Astra can now replicate complex layouts, colors, and fonts directly from a single uploaded screenshot. Front-end engineers can use this to automate UI generation and match existing design systems perfectly. If you build design tools or website builders, this is a massive capability upgrade that saves hours of manual CSS work.

Manage your Ultrafast token allowance carefully

The new Ultrafast mode generates tokens 4.5 times faster but burns through your usage limits much quicker. Pro users paying $200 a month get this speed for Sol. Those on the $500 a month Pro plan get it for both models. You trade efficiency for raw speed, so developers must use the model picker wisely when latency is critical.

My take

OpenAI is making it very hard for competitors to compete on price. Dropping Astra's cost by 20 percent while making Sol almost as smart at a fifth of the price is a brutal combination. If you are still paying premium rates for basic coding tasks, you are burning cash for absolutely no reason. I will be routing almost everything to Sol starting today.

Original reporting: 9to5Mac. This is my rewrite and opinion.

More AI news for builders