GPT-6 Astra pricing
GPT-6 Astra API usage is billed by token category and service mode; ChatGPT subscriptions and API usage are separate budgets.
Two different GPT-6 bills
GPT-6 pricing depends on which product you are using. ChatGPT is a subscription product with plan-specific model allowances. The API is a developer service billed to an API organization by usage. An active ChatGPT subscription does not automatically pay for API requests, and an API balance does not unlock a ChatGPT plan. OpenAI’s ChatGPT Help guidance also says that GPT-6 Pro availability and limits vary by plan, product and workspace rollout.
For a product decision, first identify the surface, account and allowance. The access guide covers that check. For the API model ID, endpoints and setup, see the GPT-6 API guide. This page covers the published text-token rates for gpt-6-astra.
GPT-6 Astra API rates
The official model page lists prices per 1 million tokens (USD). “Input” is uncached input; “cached input” is a cache read; “cache writes” are billed when content is written to the prompt cache; “output” is billable generated output, including reasoning tokens when they are counted in output usage.
| Processing mode | Input | Cached input | Cache writes | Output |
|---|---|---|---|---|
| Standard | $10.00 | $1.00 | $12.50 | $50.00 |
| Batch / Flex | $5.00 | $0.50 | $6.25 | $25.00 |
| Fast | $20.00 | $2.00 | $25.00 | $100.00 |
Batch and Flex are published at 50% of Standard rates. Fast is published at 2× the applicable rates. These are pricing multipliers, not a promise that every account, endpoint or combination is eligible. Confirm the mode and current account terms before committing spend; this guide does not infer unverified combinations of modes.
Caching has two separate events. A cache write is not a cache read: the first eligible request can pay the write rate, while a later eligible request can use the cheaper cached-input rate. Savings depend on what the API actually marks as cached and on repeated eligible content; they are not guaranteed for every request. Count uncached input, cache writes, cache reads and output separately in your usage records.
Long-context pricing
When a request has more than 272K input tokens, the model page says to apply 2× to input and cache rates and 1.5× to output for the full request. At Standard rates, that means $20 per 1M uncached input, $2 per 1M cached input, $25 per 1M cache writes and $75 per 1M output for that request. Do not apply the long-context rule only to tokens above 272K: the published rule applies to the full request.
Three fixed estimates
These examples use Standard rates, USD, and token counts supplied below. They are arithmetic illustrations, not a bill quote.
- 10,000 uncached input + 2,000 billable output:
(10,000 / 1,000,000 × $10) + (2,000 / 1,000,000 × $50) = $0.10 + $0.10 = $0.20. - 20,000 uncached input + 80,000 cached input + 2,000 billable output:
(20,000 × $10 + 80,000 × $1 + 2,000 × $50) / 1,000,000 = $0.20 + $0.08 + $0.10 = $0.38. This assumes an existing eligible cache and excludes any earlier cache-creation charge. - 300,000 uncached input + 10,000 billable output, long-context request:
(300,000 / 1,000,000 × $20) + (10,000 / 1,000,000 × $75) = $6.00 + $0.75 = $6.75.
The output figures in these examples are the total billable output counts and include any reasoning tokens counted in that output. Actual usage may also include tool-call fees, retries, other endpoints, taxes or account-specific terms. These examples exclude those additional items and do not include a paid API test. Review the live pricing page and your project usage before setting a customer price.
Practical budget check
Record the model, processing mode, input category, output category, token counts and request outcome. Add headroom for retries and tool calls. Keep API billing separate from ChatGPT plan allowances, and recheck official rates when the model page changes. The ChatGPT 6 guide explains the product-side terminology and plan distinction.