the Ledger

GPT-5.4 Pro

OpenAI4 benchmarks4 reviewed scoresLatest test Aug 7, 2026

Every score this site tracks for GPT-5.4 Pro, including the benchmarks that are not on the board’s headline columns. Each carries the source that shows the number, the date that source reports, and whether a curator checked it. Where a source published a margin or a range, it is shown beside the score rather than rounded away.

All tracked scores

GPQA Diamond (%)
94.6%✓
GPT-5.4 Pro GPQA Diamond source: lmcouncil.ai (opens in a new tab)Jul 1, 2026Single sourceReviewed by agent-ops
Conditions: Independent run published by LM Council; xhigh effort; ±1.6 as published
Humanity's Last Exam (%)
44.3%✓
GPT-5.4 Pro Humanity's Last Exam source: lmcouncil.ai (opens in a new tab)Jul 1, 2026Single sourceReviewed by agent-ops
Conditions: Independent run published by LM Council; ±2.0 as published
Input Cost ($/1M tokens)
30 $/1M tokens✓
GPT-5.4 Pro Input Cost ($/1M tokens) source: developers.openai.com (opens in a new tab)Aug 7, 2026Single sourceReviewed by agent-ops
Conditions: API list price for input tokens, observed 2026-08-07
Output Cost ($/1M tokens)
180 $/1M tokens✓
GPT-5.4 Pro Output Cost ($/1M tokens) source: developers.openai.com (opens in a new tab)Aug 7, 2026Single sourceReviewed by agent-ops
Conditions: API list price for output tokens, observed 2026-08-07

Method: curated scores are sourced, dated and reviewed · ✓ marks a reviewed score · unreviewed scores are scraped and labelled as such

GPT-5.4 Pro

All tracked scores

GPQA Diamond (%)

Humanity's Last Exam (%)

Input Cost ($/1M tokens)

Output Cost ($/1M tokens)