OpenAI said on September 9, 2026 that GPT-6 Astra is available in ChatGPT Work, Codex and the company API. The model is positioned for computer use, browsing, professional work, software engineering, cybersecurity and science tasks.
The company described the release as a way to run AI inside existing business software without first rebuilding data pipelines or writing custom integrations. In ChatGPT Work and Codex, the model can write code and operate the same applications employees already use, including applications that expose no API.
Pricing starts at $10 per million input tokens and $50 per million output tokens. OpenAI said the model was trained to finish tasks in fewer tokens and with fewer retries, and it cited results on Terminal-Bench 4.0 and the Artificial Analysis Intelligence Index version 4.3 to support its cost-efficiency claim.
Several software companies published early evaluation figures alongside the launch. Cognition integrated the model into its Devin harness on launch day. Databricks reported better cost per task than the earlier GPT-5.6 Sol model on its OfficeQA Pro benchmarks. Hebbia measured 17 percent closer adherence to a written brief and 19 percent more accurate document sourcing than the next best model it tested. CodeRabbit reported about 20 percent more bugs caught in code review. Datacurve recorded 74 percent on DeepSWE version 1.1.
OpenAI also described internal use of the model. Its engineering team traced a memory allocation bottleneck in a Codex test environment and reported 25 times lower turn latency after switching allocators, with roughly 30 percent higher peak memory use.
Source: OpenAI - https://openai.com/index/gpt-6-astra-next-generation-work