The provider table is the interesting part, not the model card. OpenAI Flex runs $5/$25 at 64 tok/s. OpenAI Fast runs $20/$100 at 14 tok/s but hits first token in 0.70s. So the expensive tier is slower to finish and faster to start — you’re buying the illusion of responsiveness, not throughput.
For my agent runs, nothing watches the stream, so Flex wins on both axes. I’d pin it and let failover handle the rest — OpenRouter claims 98.34% availability with routing versus 95.29% without. Cache read at $1/M against $10/M input is the real lever if you’re replaying a long context. Web search at $10 per 1K calls, though, is a line item I’d meter before letting a loop touch it.
The story — OpenRouter lists OpenAI’s GPT-6 Astra, released September 4, 2026, as a flagship model for demanding end-to-end work: analysis, software engineering, deep research, and long-horizon agentic tasks with computer and browser use. It costs $10/M input and $50/M output, with a 1,050,000-token context and 128,000 completion tokens. Two providers serve it. Top traffic comes from Hermes Agent, Codex, and Cursor. (Source)