OpenAI's GPT-6.2 Astra Beats Its Predecessor at Half the Cost per Task
Independent testers at Vals AI clock the new flagship at $10.68 per task, down from $21.28, while API prices drop 20%.

OpenAI's new flagship finishes the same professional task for about half what its predecessor charged: $10.68 against $21.28, according to independent testers at Vals AI. GPT-6.2 Astra and its cheaper sibling, GPT-6.2 Sol, went live on November 17 in ChatGPT Work, Codex and the API. For the developers and companies that pay per token, it is the rare upgrade that scores higher and bills lower at the same time.
Astra for the Hard Jobs, Sol for Everything Else
The two models split the work. OpenAI pitches Astra as a frontier model "built for your hardest projects", with science, law and finance among them, and Sol as an efficient, versatile model for everyday coding and writing. Both arrive in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and in the API as gpt-6.2-astra and gpt-6.2-sol.
The company announced them on X with a wink at its impatient users.
“Some of you have been asking very politely. Some of you have just been asking.”— OpenAI
Speed got an upgrade too. Pro 500 subscribers can run either model in Ultrafast mode, and Pro 200 subscribers get Ultrafast for the first time, starting with Sol. Enterprise and Edu admins switch the models on for their teams, and regular ChatGPT chat keeps its current models for now.
So how much better is Astra, and what does the lower bill actually buy?
From $21.28 to $10.68 a Task
API prices fell 20%: Astra now costs $8 per million input tokens and $40 per million output tokens, while Sol keeps its $2 and $10. The bigger saving comes from efficiency. On GDPval-v2, OpenAI's test of real professional work, Astra gains 6.2 points over its predecessor while writing 22% fewer output tokens.
OpenAI says Astra sets its new highs "at less than a third of the cost per task of GPT‑6 Astra," measured on its own agent and science benchmarks. Vals AI, which runs an independent index, found a smaller gap that is still worth a CFO's attention.
The scores moved up across the board. On Agents' Last Exam, Astra climbs from 73.1% to 77.6%. On Terminal-Bench-Science it jumps from 68.1% to 77.1%, a nine-point gain. Outside OpenAI's own charts, Artificial Analysis ranks Astra second on its Intelligence Index with 62 points, behind Anthropic's Claude Fable 5.2 at 65, and Epoch AI's Capabilities Index puts the two level at 174.
Astra is the headline. The model that may matter more to most developers is the cheaper one.
Sol Nearly Matches Astra at a Fifth of the Price
On FrontierCode 2.0, which grades merge-ready changes to real codebases, Sol scores 43.8% against Astra's 44.2%, at about one-fifth of the cost, according to OpenAI. On the Vals index, Sol runs at $3.18 per task. For a startup running thousands of coding agents a day, near-flagship quality at that price is the difference between a pilot and a product.
The new capability came with tighter rules. Both models reach the "Critical" cybersecurity level in OpenAI's Preparedness Framework, as GPT-6 Astra did, so they launch with stricter safeguards on higher-risk security help. OpenAI says it is working to bring them to Daybreak, so approved security defenders get fewer refusals.
The biggest winners are the people who pay by the task: a lab running analyses overnight, or a developer whose coding agent now costs half as much to keep running. Half the cost per task means twice the work for the same budget, starting today.
What people are saying
“Some of you have been asking very politely. Some of you have just been asking. GPT-6.2 Astra and Sol are now available. Super efficient and built to perform where it counts—the real world.”
Higher Scores, Half the Bill
The Brief
Stay curious
AI and technology: what changes and why it matters.
Your daily selection, in English or Spanish.
Free forever. Unsubscribe anytime.
More stories

The Specialty News





Conversation
Start the conversation