GPT-4o
Last updated: August 10, 2026
Specs
| Vendor | OpenAI |
|---|
| Released | May 13, 2024 |
|---|
| Context window | 128K tokens |
|---|
| Input | $2.5 / M tokens |
|---|
| Output | $10 / M tokens |
|---|
Benchmarks
| mmlu | 88.7 |
|---|
| gpqa | 53.6 |
|---|
| humanEval | 90.2 |
|---|
| arenaElo | 1345 |
|---|
| sweBench | 33 |
|---|
| sweBenchPro | 22 |
|---|
| liveCodeBench | 50 |
|---|
| aiderPolyglot | 45 |
|---|
| terminalBench | 18 |
|---|
| mmluPro | 72 |
|---|
| hle | 5 |
|---|
| aime | 42 |
|---|
| math500 | 76.6 |
|---|
| mmmu | 69 |
|---|
Strengths
- Most battle-tested multimodal model (text+vision+audio)
- Vast ecosystem and tool-calling support
- Stable and widely integrated
Weaknesses
- Superseded by GPT-5 on reasoning and coding
- Smaller context than GPT-5 (128K vs 400K)
- Less sharp on complex math
Best for
- Existing GPT-4o pipelines
- Realtime voice applications via the Realtime API
- Cost-conscious multimodal apps
Head-to-head comparisons
Compare with other AI models