OpenAI releases GPT-6.2 Astra and Sol with lower costs
OpenAI’s new frontier pair targets professional agents and everyday workflows, combining higher benchmark scores with cheaper inference and tighter safeguards.
OpenAI has released GPT-6.2 Astra and GPT-6.2 Sol, dividing its latest model generation between demanding professional work and lower-cost everyday tasks. Both models are available immediately in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, as well as through the OpenAI API.
A stronger professional model at a lower price
OpenAI says Astra improves long-horizon knowledge work, coding, science, document production and computer use. On the company’s Agents’ Last Exam, it scored 5.2 points above GPT-6 Astra while costing less than one-third as much per task. OpenAI also reports a nine-point gain on Terminal-Bench-Science 0.1 and first place on its AA-3D evaluation, where human judges preferred Astra-built interactive experiences over competing outputs.
The API launch is economically important. Astra’s standard price is $8 per million input tokens and $40 per million output tokens, with cached input priced at $0.40. OpenAI says those rates represent a 20% reduction from GPT-6 Astra, while improved token efficiency lowers the total cost of longer workflows further.
Sol broadens agent deployment
Sol is positioned as the default model for routine multistep work. OpenAI says it comes within 0.5 points of Astra on Cognition’s FrontierCode 2.0 while operating at roughly one-fifth of the cost. That combination makes it more relevant to large-scale agent deployment than a pure flagship benchmark win.
The release also raises the safety bar. OpenAI classifies both models as reaching its Critical threshold for cybersecurity and High threshold for biological and chemical capabilities. Astra additionally reaches the High threshold for AI self-improvement, prompting tighter restrictions, enhanced monitoring and staged access for higher-risk use cases.
Why it matters
This is more than a routine model refresh: OpenAI is pairing stronger autonomous work with materially lower unit economics. If the reported efficiency survives independent testing, the competitive pressure will shift from who has the best model to who can run capable agents continuously, safely and profitably.