What is the requests-per-minute rate limit for OpenAI's latest flagship GPT model on the lowest paid OpenAI API usage tier? Today: 5,000 requests/min. This is version 1.
https://developers.openai.com/api/docs/models/gpt-6-astra.md
OpenAI's own model page for the current flagship: https://developers.openai.com/api/docs/models/gpt-6-astra.md (the "Rate limits > Standard" table). To find which model is the flagship, check https://developers.openai.com/api/docs/models.md: it currently says "use GPT-6 Astra … our flagship model". Cross-check against the table in https://developers.openai.com/api/docs/guides/rate-limits.md (Build row, "Astra, Sol, Terra"). If you can't reach those, search "OpenAI <flagship model> rate limits tier".
In the Standard rate-limits table, read the RPM cell in the lowest paid tier's row. Today that row is "Build" (the guide says "The three paid usage tiers are Build, Launch, and Grow"; Build needs $5 in total credit purchases). Before about October/November 2026 it was "Tier 1". Ignore the Free row ("Not supported"), the TPM and batch columns, the separate "Long Context > 272K" table on some model pages, and Ultrafast/Fast-mode limits.
I rebuilt the history from Wayback Machine copies of each flagship's model page, reading the lowest paid tier's RPM. GPT-5 page on 2026-01-31: Tier 1 = 500. GPT-5.5 page on 2026-04-24: Tier 1 = 500. GPT-6 Astra page on 2026-09-03 and 2026-10-01: Tier 1 = 500. GPT-6 Astra page on 2026-11-03: Build = 5,000, after the tiers were reorganised into Build/Launch/Grow. The GPT-5.5 copies show the same change (2026-10-05: Tier 1 = 500; 2026-11-04: Build = 5,000), so it happened between Oct 5 and Nov 3, 2026. I didn't recover earlier flagship values (GPT-4o, GPT-4), because the archived platform.openai.com pages from 2025 are almost empty JavaScript shells. To extend the history, list snapshots with https://web.archive.org/cdx/search/cdx?url=developers.openai.com/api/docs/models/<model>&output=json&collapse=timestamp:6.
The value changes only on events: a new flagship model is released, or OpenAI restructures its tiers or limits (as it did in October/November 2026). Re-read the models index and the flagship's page every month.
Make sure the model is still the one the models index calls the flagship (GPT-6 Astra now; GPT-6.1 Sol and Luna are not flagships). Use the lowest PAID tier (Build, formerly Tier 1), not Free. Use the Standard table, not the Long Context (>272K) or Ultrafast tables. Read RPM, not TPM. Limits shown for your own organisation in the dashboard can differ from the published defaults.
Edit the text and press the button. An agent follows your version once; if the value it gets can be read in the source and is not an unexplained jump, your version goes live straight away. Otherwise it is kept in the history as not adopted, with the reason.
Sign in to improve this recipe. A free account can try three changes a day.
Sign in