Long-context pricing

GPT-6 Astra 272K pricing cliff

OpenAI describes higher rates for prompts with more than 272K input tokens. This page explains how the unofficial calculator models that cliff.

Trigger we use

Cliff applies when:

(uncached_input + cached_input) > 272000

Cache write tokens are billed separately and are not added into the cliff threshold. The threshold follows “prompt input tokens” language: uncached + cached input.

Multipliers on the full request

Input rate × 2
Cached input rate × 2
Cache write rate × 2
Output rate × 1.5

Once the cliff triggers, multipliers apply to the entire request bill — not only the tokens past 272K. After cliff rates are set, Batch/Flex (0.5×) or Fast (2×) still apply.

Quick example

200K uncached + 100K cached = 300K input → cliff on. Standard effective rates become input $20, cached $2, cache writes $25, output $75 per 1M. With 20K output and no cache writes:

Input: 200K × $20/1M = $4.00
Cached: 100K × $2/1M = $0.20
Output: 20K × $75/1M = $1.50
Total ≈ $5.70 per request

Try it live on the homepage calculator (auto-detect is on by default).

Verify officially

This explanation is unofficial. Read OpenAI’s wording on GPT-6 Astra and API pricing before relying on these numbers.