AI News

OpenAI o3 cache-read price history

Kingy.ai Price History Library · evidence observed through August 2, 2026

OpenAI o3 cache-read price history tracks cache read list pricing for o3-2025-04-16 (served by the o3 alias). It publishes 1 editorially verified episode while preserving the broader machine timeline as a separate evidence layer.

OpenAIprovider
o3-2025-04-16 (served by the o3 alias)stable identity
cache readprice component
USD / 1M tokensunit
42retained observations
2machine-detected states

Identity and scope

The April 2025 o3 API release is the o3-2025-04-16 snapshot, also served through the o3 alias. OpenAI described the June change as the same exact model becoming cheaper, which rules out a replacement-model explanation for this episode.

Scope boundary. Values are published list prices in USD per million native tokens for the named component. They do not establish negotiated, enterprise, regional, cloud-marketplace or promotional rates. The archive records when a price was observed, not the exact moment a provider changed billing.

Editorially verified change

Cached input fell from $2.50 to $0.50 per million tokens. The 80% reduction preserves the 25% relationship between cached and uncached input in both observed price states.

Old price last seen Old New price first seen New Verification
June 8, 2025
last old-price capture
$2.5 June 11, 2025
first new-price capture
$0.5 80% cut; OpenAI announced the change June 10 and the retained pricing capture observed it by June 11 UTC.

Machine-detected state timeline

The rows below come from the scope-safe machine series. A changed value is not automatically an editorial change. Only rows tied to the verified brackets above support the page’s change claim; the others remain observations or held transitions.

State First observed Value Editorial status Lineage
1 $2.5 USD machine-only state archive
6b4dd816f708e8ad…
2 $0.5 USD bounds a verified episode archive
fbf62536bcf75db7…

Source and archive lineage

How to use this history

Use the verified row when reconstructing a dated standard-list-price estimate. Use the machine timeline to audit extraction and locate captures, not to infer billing scope or a precise effective timestamp. Recheck the provider’s current documentation before making a live purchasing or architecture decision.