GPT-5
Last updated: August 10, 2026
Specs
| Vendor | OpenAI |
|---|
| Released | August 7, 2025 |
|---|
| Context window | 400K tokens |
|---|
| Input | $1.25 / M tokens |
|---|
| Output | $10 / M tokens |
|---|
Benchmarks
| mmlu | 91.5 |
|---|
| gpqa | 63.8 |
|---|
| humanEval | 91.6 |
|---|
| arenaElo | 1412 |
|---|
| sweBench | 65 |
|---|
| sweBenchPro | 56 |
|---|
| liveCodeBench | 75 |
|---|
| aiderPolyglot | 76 |
|---|
| terminalBench | 45 |
|---|
| mmluPro | 85 |
|---|
| hle | 25 |
|---|
| aime | 92 |
|---|
| math500 | 96 |
|---|
| mmmu | 78 |
|---|
Strengths
- Balanced reasoning, coding and multimodal
- Mature tool-calling ecosystem
- Native audio I/O
Weaknesses
- Behind Claude Opus 4.7 on long-context coding
- Rate limits on the highest tiers can be restrictive
Best for
- Multimodal assistants (voice + text + image)
- General-purpose agents
- Retrieval-augmented generation
Compare with other AI models