THE CRUNCH

OpenAI has released two new language models, GPT-6.2 Astra for demanding professional work and GPT-6.2 Sol for everyday use, and says both beat their predecessors on benchmarks while costing less to run. Both are available now in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and developers can call them through the OpenAI API. Pro-500 subscribers get Ultrafast mode with both models, and Pro-200 subscribers get Ultrafast access for the first time, starting with Sol.

API pricing for Astra is 20 percent below its predecessor's, at $8 per million input tokens, $0.40 for cached inputs and $40 per million output tokens. Sol keeps GPT-6.1 Sol's pricing of $2 for input, $0.10 for cached inputs and $10 for output per million tokens. The launch continues a price-cutting run: in late October OpenAI released GPT-6.1 Terra, a mid-range model that halved API prices compared with its predecessor.

The headline benchmark numbers come from OpenAI's own testing. The company says Astra scores 5.2 points above GPT-6 Astra on Agents' Last Exam, a test of multi-step professional workflows, at less than a third of the cost, and more than doubles Claude Fable 5.2's pass rate on a legal benchmark at about one-sixth of the cost. Sol trails Astra by just 0.5 points on the FrontierCode 2.0 coding benchmark at one-fifth the cost, according to OpenAI. Independent figures are thinner: on the Artificial Analysis Intelligence Index, which rolls up nine evaluations, Astra finishes second behind Claude Fable 5.2, and Artificial Analysis says OpenAI holds most of the top spots for cost-versus-performance efficiency.

OpenAI says both models hit the "Critical" threshold for cybersecurity capability in its Preparedness Framework, meaning they find and exploit vulnerabilities more effectively, and is adding stricter safeguards for potentially risky cybersecurity requests. It notes these will also limit legitimate vulnerability research. The published cybersecurity benchmark numbers come from internal tests with safeguards and classifiers switched off, with methodology laid out in an accompanying safety report.

WHAT HAPPENS NEXT

OpenAI plans to offer the new models through OpenAI Daybreak, a programme that gives vetted security researchers access to models with fewer guardrails.