GPT-5.4 and GPT-5.4 Pro
GPT-5.4 is OpenAI's first general-purpose model with native computer use, scoring 75.0% on OSWorld-Verified, with a 1M-token context.
Folds the Codex coding line into the flagship, adds tool search and compaction in the API, and improves spreadsheets, slides and deep research. GDPval 83.0% wins or ties vs professionals. OSWorld-Verified 75.0% versus 47.3% for GPT-5.2.
- Date
- Thursday, 5 March 2026
- Lab
- OpenAI
- Kind
- model
- Access
- closed API
- Price
- GPT-5.4 $2.50/$15 per 1M tokens (mini $0.75/$4.50, nano $0.20/$1.25 added 2026-03-17)
Figures
| Measure | Value | Measured by |
|---|---|---|
| OSWorld-Verified | 75.0% GPT-5.2 47.3%; OpenAI cites a 72.4% human baseline | company |
| GDPval (wins or ties) | 83.0% GPT-5.2 70.9% | company |
| SWE-Bench Pro (public) | 57.7% GPT-5.3-Codex 56.8% | company |
| Claim-level factual errors | 33% fewer than GPT-5.2 18% fewer false responses; flagged-prompt set | company |
Knowledge cutoff 2025-08-31 (Simon Willison). Flagship price from his 2026-03-17 comparison table.
Sources
- openai.com/index/introducing-gpt-5-4/
- deploymentsafety.openai.com/gpt-5-4-thinking/
- simonwillison.net/2026/Mar/5/introducing-gpt54/
- developers.openai.com/api/docs/changelog
This record was checked against its sources on 6 October 2026. How we check