GPT-5.4
March 2026 frontier model with 33% fewer factual errors and built-in computer use.
About
Released March 5, 2026 (Thinking and Pro tiers; mini and nano followed March 17), GPT-5.4 is OpenAI's frontier model for professional knowledge work, reasoning, coding, and agents. OpenAI reported a 33% reduction in factual errors versus GPT-5.2, built-in computer-use capabilities, and improved deep research. On OSWorld-Verified — a benchmark of desktop computer use — GPT-5.4 scored 75%, versus 47.3% for GPT-5.2 and 72.4% for the average human. ZDNET praised its low hallucination tendency; Vice noted improved deep research with fewer tokens than GPT-5.2 Thinking.
Limitations: the mini and nano variants cost up to four times more through the API than their GPT-5 equivalents, which The Decoder criticized. ZDNET found GPT-5.4 Thinking sometimes fails to follow user prompts accurately despite strong answers. Thinking and Pro were not available to free-tier users at launch.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
OpenAI-reported: 75% on OSWorld-Verified (vs 47.3% GPT-5.2, 72.4% human); 33% fewer factual errors than GPT-5.2
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 56; Trending score 88/100 from researcher assessment.