OpenAI Upgrades GPT-6 Prompt Caching with Dashboard and Up to 90% Input Savings

On September 22–23, OpenAI announced an upgrade to prompt caching for GPT-6 in its API, making long-running agents faster and cheaper, backed by new diagnostics tooling. For developers and agent applications that frequently call with long contexts, this amounts to a substantive price cut.

Confirmed

Default cache hit rates are higher, so more input tokens qualify for up to 90% cached-input discounts; @rhiever reports cached input rates were also lowered. Per billing rules compiled by @新智元, for GPT-5.6 and later models cached writes cost 1.25x normal input while cache hits read at 0.1x, saving roughly 78.5% on repeated material.
The improvements directly cut costs for repeated context such as long system prompts and conversation history.
A Prompt Caching Dashboard launched on the platform, letting developers see what is reused, what changes break hits, and use a diagnostics API to pinpoint failures.
New capabilities include explicit cache breakpoints to choose which prompt prefixes to reuse, adjusting reasoning effort and tool availability while preserving cached context, and prewarming shared context for faster responses.
@rohanpaulai notes the update makes caching more reliable and controllable, including inference behavior changes that preserve the cache—especially valuable for long-running agents maintaining state across calls.

Why it matters

Repeatedly sending identical prefixes is a major cost driver in long-context and agent workflows; the 1.25x write / 0.1x read pricing with higher hit rates can significantly reduce running costs.
The dashboard and breakpoints turn caching from luck into an observable, controllable engineering practice.

2026-09-23 ~ 2026-09-24 · 6 related posts

Primary sources

OpenAI Boosts GPT-6 Prompt Caching, Input Tokens Now Up to 90% Cheaper — OpenAIDevs · 2026-09-23 05:13 (UTC+08:00)
OpenAI Launches Prompt Caching Dashboard to Track Cache-Hit Rates and Misses — OpenAIDevs · 2026-09-23 05:13 (UTC+08:00)
OpenAI upgrades GPT-6 prompt caching: 0.1x read pricing, up to 78.5% savings on repeated inputs — 新智元 · 2026-09-24 07:33 (UTC+08:00)

[source] OpenAI Boosts GPT-6 Prompt Caching, Input Tokens Now Up to 90% Cheaper — OpenAIDevs · 2026-09-23
[source] OpenAI Launches Prompt Caching Dashboard to Track Cache-Hit Rates and Misses — OpenAIDevs · 2026-09-23
GPT-6 prompt caching gets more reliable: higher hit rates, diagnostics, up to 90% savings — rohanpaul_ai · 2026-09-23
OpenAI rolls out upgraded prompt caching and lower cached input rates for GPT-6 — rhiever · 2026-09-24
[source] OpenAI upgrades GPT-6 prompt caching: 0.1x read pricing, up to 78.5% savings on repeated inputs — 新智元 · 2026-09-24

1 near-duplicate retellings: OpenAIDevs