AI models
DeepSeek V3 vs GPT-5
Same criteria, side by side. Where the gap is small, it may be measurement noise.
Also searched as
DeepSeek V3 vs GPT-5
Direct answer: GPQA Diamond: DeepSeek V3 67.6%, GPT-5 86.2% (Epoch AI). OTIS Mock AIME: DeepSeek V3 37.8%, GPT-5 91.4% (Epoch AI). GPQA Diamond (OpenRouter): DeepSeek V3 66.7%, GPT-5 85.7% (OpenRouter Benchmarks). τ-bench Verified (airline): DeepSeek V3 43.8%, GPT-5 70.7% (OpenRouter Benchmarks).
| Metric | DeepSeek V3 | GPT-5 | Source |
|---|---|---|---|
| Weights ?what it is | open | closed | no data |
| Input US$/1M ? | US$ 0.29 | US$ 1.25 | no data |
| Output US$/1M ? | US$ 1.14 | US$ 10.00 | no data |
| Context ? | 163,840 | 400,000 | no data |
| GPQA Diamond ? | 67.6% | 86.2% | Epoch AI |
| OTIS Mock AIME ? | 37.8% | 91.4% | Epoch AI |
| FrontierMath (tiers 1 to 3) ? | no data | 55.4% | Epoch AI |
| SimpleQA Verified ? | no data | 50.1% | Epoch AI |
| SWE-bench Verified ? | no data | 73.6% | Epoch AI |
| GPQA Diamond (OpenRouter) ? | 66.7% | 85.7% | OpenRouter Benchmarks |
| τ-bench Verified (airline) ? | 43.8% | 70.7% | OpenRouter Benchmarks |
Sources
- Epoch AI, CC BY 4.0, latest version we checked.
- OpenRouter Benchmarks, CC BY 4.0, latest version we checked.
- OpenRouter, dados públicos da API; exports de benchmarks e rankings em CC BY 4.0, latest version we checked.
Dates come from the sources: OpenRouter listing or Hugging Face repo creation, which may differ from the official launch. Family and version were grouped by rule from the name.