GPT-5

published

OpenAI's frontier model

OpenAI's most capable model with native tool use and extended reasoning. 94.6% on AIME 2025 without tools.

Provider openai Type llm Access api Params undisclosed Context 128k License proprietary

Benchmarks (5)

Benchmark	Score	Reported By	Source
AIME 2025	94.6%	vendor	source ↗
SWE-bench Verified	74.9%	vendor	source ↗
Aider Polyglot	88.0%	vendor	source ↗
GPQA (Pro)	88.4%	vendor	source ↗
MMMU	84.2%	vendor	source ↗

Massive step up in mathematical reasoning (94.6% AIME). GPT-5 Pro extended reasoning reaches 88.4% GPQA without tools.

High cost per token. No open weights. Training data cutoff less recent than competitors.

Provider openai

Released 2025-12-01

Training cutoff 2025-10

Created March 22, 2026

Last reconciled Never