How many credits does GPT-6 Astra cost in ChatGPT?
On OpenAI's rate card, GPT-6 Astra costs 250 credits per million input tokens, 25 cached and 1,250 per million output. A GPT-6 Pro chat message is 50 credits, and Fast mode costs 2.5 times the standard rate.
Checked against the lab's own pages on September 18, 2026. Limits change often, so the date matters.
OpenAI's credit rate card is where the GPT-6 Astra numbers live, and it splits them in two. Chat is priced per message. Work and Codex are priced by the token, which is why two people on the same plan report wildly different burn.
Astra is the most expensive row on the token table, and output costs five times input.
- GPT-6 Astra input
- 250 credits per 1M tokens
- Cached input
- 25 credits per 1M tokens
- Output
- 1,250 credits per 1M tokens
- GPT-6 Pro chat message
- 50 credits
- Fast mode
- 2.5 times the Standard rate
- GPT-6 Astra Law input
- 312.50 credits per 1M tokens
What a chat message costs
GPT-6 Pro is powered by GPT-6 Astra, and a GPT-6 Pro chat message costs 50 credits, the same rate as GPT-5.6 Sol Pro. Chat usage is metered separately from Astra usage in Work and Codex, so the two don't come out of one bucket.
For scale on the same card, GPT-5.6 Sol is 10 credits a message and Instant is unlimited. Picking Medium, High or Extra High doesn't move the rate, because all three run GPT-5.6 Sol.
What Work and Codex cost
Astra runs 250 credits per million input tokens, 25 per million cached input, and 1,250 per million output. GPT-5.6 Sol on the same table is 100, 10 and 500 (input, cached, output, in that order). Luna is 5, 0.5 and 30, which puts Astra at fifty times Luna on input.
So the shape of the task decides the bill more than the length does. Granted, the card says actual usage also moves with task complexity, automations and the number of concurrent Codex instances.
Output-heavy tasks and fast mode generally consume more credits than lighter tasks.
Fast mode is the multiplier to watch. For GPT-6 Astra in Work and Codex it's charged at 2.5 times the Standard rate, and Codex doesn't charge for cache writes at all.
What happens when the included usage runs out
On ChatGPT Business, standard seats use the included plan limits first, and eligible usage then continues from the shared credit pool if the workspace has bought credits. Codex-only seats are the exception. They include Codex usage only, and they need workspace credits from first use (no included allowance to burn through first).
Enterprise and Edu workspaces on credit-based flexible pricing scale usage with credits and have no fixed rate limits. Workspaces without flexible pricing stay on their plan limits.
Who this card covers
Business and Enterprise/Edu workspaces on credit-based flexible pricing. If an Enterprise agreement bills in US dollars, a different rate card applies, and Plus and Pro rates sit in a separate article that OpenAI links from this one. The agreement decides which one governs.
One date on the card has gone stale. It still lists the Codex retirement of GPT-5.4 and GPT-5.4 mini on Aug 31, 2026 as upcoming, for users signed in with ChatGPT. The live one is GPT-5.6 Sol's promotional pricing, which the card says runs at least through November 21, 2026.
Read on September 18, 2026, when the page said it had been updated 24 hours earlier. Rates move.