Article · AI translation
Thanks to IT Home readers Domado, JARK006, 阿狗, SoftMedia user 1773925, 84路龚娇霉, 我家小妞很拽, 洛書, HuTao29, new SoftMedia user 1916289, Nextpage, new SoftMedia user 2205747, 华南吴彦祖, 兰花是我, and SoftMedia user 1053564 for the tips!
IT Home reported on November 18 that OpenAI officially released its latest GPT-6.2 Astra and GPT-6.2 Sol models today, focusing on improving performance in professional work, programming, computer use, and multistep agent tasks while reducing the number of output tokens needed to complete tasks.
OpenAI describes GPT-6.2 Astra as its flagship model for complex professional projects, with further improvements in professional work, scientific research, software development, and visual design. The model can follow user-provided presentation templates more accurately, improve the quality of slides, spreadsheets, and documents, and enhance its capabilities in frontend development, website design, game development, and 3D design.
In the Agents' Last Exam evaluation, GPT-6.2 Astra scored 5.2 percentage points higher than the previous-generation GPT-6 Astra, with a cost per task of less than one-third as much. In the Terminal-Bench-Science 0.1 scientific research evaluation, the new model scored 9 percentage points higher than GPT-6 Astra, also at less than one-third the cost per task of the previous generation.
OpenAI also said GPT-6.2 Astra set a new high on the Vals Science Index evaluation. In Artificial Analysis's AA-3D evaluation, the model ranked first with an Elo rating of 1318, ahead of Claude Fable 5.2 at 1271 Elo and GPT-6.2 Sol at 1259 Elo.
The other model, GPT-6.2 Sol, is positioned for everyday work and balances efficiency with general-purpose capabilities. Compared with GPT-6.1 Sol, the new model improves on agent workflows, coding, and handling multistep tasks, can complete tasks using fewer tokens, and produces more natural writing.
OpenAI said that in Cognition's FrontierCode 2.0 evaluation, GPT-6.2 Sol scored within 0.5 points of GPT-6.2 Astra at about one-fifth the cost per task. In the LifeVal everyday-task evaluation, GPT-6.2 Sol outperformed GPT-6 Astra at about one-seventh the cost per task.
On pricing, the standard GPT-6.2 Astra API rates are $8 per million input tokens, $0.4 per million cached input tokens, and $40 per million output tokens. Compared with GPT-6 Astra, the new model's input and output token prices are both 20% lower, while its cached input token price is 60% lower.
GPT-6.2 Sol retains GPT-6.1 Sol's API pricing: $2 per million input tokens, $0.1 per million cached input tokens, and $10 per million output tokens.
On availability, OpenAI said both models became available on November 17 to Plus, Pro, Enterprise, and Edu users of ChatGPT Work and Codex. Pro 500 users can use Ultrafast mode with both models, while Pro 200 users gain access to GPT-6.2 Sol's Ultrafast mode for the first time. Eligible Enterprise and Edu users also need a workspace administrator to enable the relevant features.
Developers can also access gpt-6.2-astra and gpt-6.2-sol through the OpenAI API, with Ultrafast mode available for both models at launch.
In addition, according to OpenAI's official release notes, GPT-6.2 Astra and GPT-6.2 Sol are primarily intended for ChatGPT Work, Codex, and the API and were not added to ChatGPT's regular chat mode at the same time. IT Home learned that the releases focus on improving the quality of complex task completion and token efficiency, allowing users to get more work done within their existing usage limits.