September 30, 2026
GPT-6 Astra generates responses faster: OpenAI claims up to 8x faster generation
Ultrafast speeds up generation with the same model, but costs 6 times as much as Standard in the API.
OpenRouter
@openrouter
Astra Ultrafast 🤯 https://openrouter.ai/openai/gpt-6-astra
· 3.7K views
GPT-6 Astra Ultrafast generates tokens in Codex up to 8 times faster than Standard, according to OpenAI as of Sep 30. This is generation speed, not task completion speed.
OpenRouter added an accelerated mode for openai/gpt-6-astra. The model stays the same; users choose the speed and price of the request.
The price of speed. In the OpenAI API, Standard costs $10 per million input tokens and $50 per million output tokens. Ultrafast is priced separately.
| Mode | Input, per 1 million tokens | Output, per 1 million tokens | | --- | --- | --- | | Standard | $10 | $50 | | Ultrafast | $60 | $300 | | Ultrafast with more than 272,000 input tokens | $120 | $450 |
Prices were checked on Sep 30, 2026 against OpenAI pricing and the GPT-6 Astra documentation. When the threshold is exceeded, the higher rate applies to the entire request. OpenRouter charges $60/$300 for Ultrafast, with cache reads costing $6 per million tokens.
How to enable it. In OpenRouter, pass `service_tier: "ultrafast"`. The `:nitro` and `:floor` suffixes do not select this mode. OpenRouter allows fallback to priority and Standard, while `provider.only: ["openai/ultrafast"]` restricts requests to Ultrafast.
In the OpenAI API, enable the mode through `client.responses.create` with `model="gpt-6-astra"` and `service_tier="ultrafast"`. The documentation includes ready-to-use examples for Python, JavaScript, and curl.
Ultrafast is available to all API users. Limits are 500,000 tokens per minute for usage tiers 1–3, 1 million for tier 4, and 5 million for tier 5. The mode supports global processing and the US region; regional endpoints in the EU and other regions outside the US are not supported.
In Codex and ChatGPT Work, the mode is available on the $500 Pro plan and eligible Enterprise/Edu plans. Included usage is consumed 8 times faster than Standard, while purchased credits are charged at a multiplier of 6.
Original source: [GPT-6 Astra on OpenRouter](https://openrouter.ai/openai/gpt-6-astra).
For AI agents that call tools frequently, OpenAI recommends a persistent WebSocket connection so network latency does not erase Ultrafast's speed advantage.
