OpenAI Launches GPT-6 Astra, Its Most Capable Model Yet

OpenAI has released GPT-6 Astra, its most capable model to date. The flagship ships with a 1.05 million token context window, up to 128,000 output tokens, and five selectable reasoning levels, and reports 96.0% on GPQA Diamond.

OpenAI Launches GPT-6 Astra, Its Most Capable Model Yet

OpenAI has released GPT-6 Astra, the new flagship of the GPT-6 family and the company's most capable model to date. The release, which began rolling out on September 3, 2026, is positioned squarely around complex agentic tasks — long-running, multi-step work where a model has to plan, use tools, and recover from errors without human intervention.

The headline specifications are a significant step up from earlier generations. GPT-6 Astra supports a 1.05 million token context window and can generate up to 128,000 tokens of output in a single response — enough to process large codebases, lengthy document sets, or entire project histories in one pass. OpenAI also introduced five selectable reasoning levels, letting developers trade latency and cost against depth of reasoning for each request.

On benchmarks, OpenAI reports 96.0% on GPQA Diamond, the graduate-level science reasoning test that has become a standard reference point for frontier models. The company describes broad gains across its evaluation suite compared with earlier models, though independent verification is still accumulating.

API pricing is set at $10.00 per million input tokens and $50.00 per million output tokens. That positions Astra at a premium of roughly 2.5x over GPT-6 Sol, the cheaper tier OpenAI introduced for cost-sensitive workloads. Third-party providers are already reselling access at slightly higher rates — Venice, for example, lists Astra at $12.50/$62.50 per million tokens.

Access is rolling out in stages, with paid API users first in line. For teams building agents, the combination of a million-token window, long outputs, and tunable reasoning makes Astra the most flexible model OpenAI has shipped — provided the budget matches the capability.