Quality
Highest overall scores
- 1
DeepSeek: DeepSeek V4 Pro44.3 - 2DeepSeek: DeepSeek V4 Flash40.3
- 3DeepSeek: DeepSeek V3.232
DeepSeek model benchmarks.
Compare DeepSeek models across answer quality, coding, complex tasks, speed, price, and context size.
Response speed
Fastest output in this sample
- 1DeepSeek: DeepSeek V4 Flash92 t/s
- 2DeepSeek: DeepSeek V4 Pro87 t/s
- 3DeepSeek: DeepSeek V3.268 t/s
Input price
Lowest price per 1M tokens
- 1DeepSeek: DeepSeek V4 Flash$0.14/1M
- 2DeepSeek: DeepSeek V3.2$0.27/1M
- 3DeepSeek: DeepSeek V4 Pro$0.43/1M
Highlights
Quality
Overall capability · Higher is better
Speed
Output tokens per second · Higher is better
Input price
USD per 1M tokens · Lower is better
Quality benchmarks
Compare DeepSeek model capability.
Performance and price
Compare practical tradeoffs.
All DeepSeek models.
3 models in the current catalogue.
| Model | Links | |||||
|---|---|---|---|---|---|---|
| DeepSeek: DeepSeek V4 Pro | 44.3 | 87 t/s | 1M | $0.43 | $0.87 | |
| DeepSeek: DeepSeek V4 Flash | 40.3 | 92 t/s | 1M | $0.14 | $0.28 | |
| DeepSeek: DeepSeek V3.2 | 32 | 68 t/s | 163.8K | $0.27 | $0.40 |
DeepSeek technical benchmarks.
Compare named evaluation records available for DeepSeek models. Missing results remain blank rather than being estimated.
| Benchmark | Area | DeepSeek V4 Pro | DeepSeek V4 Flash | DeepSeek V3.2 | Source |
|---|---|---|---|---|---|
| Intelligence Index | overallQuality | 44.3 | 40.3 | 32 | OpenRouter / artificial-analysis |
| Coding Index | coding | 59.4 | 56.2 | 44.2 | OpenRouter / artificial-analysis |
| Agentic Index | agentic | 36.4 | 31.1 | 18.3 | OpenRouter / artificial-analysis |
| GPQA | GPQA Diamond | 88.8 | 89.4 | 84 | OpenRouter model benchmarks |
| Humanity's Last Exam | HLE | 35.9 | 32.1 | 22.2 | OpenRouter model benchmarks |
| IFBench | IFBench | 76.5 | 79.2 | 60.7 | OpenRouter model benchmarks |
| τ²-Bench Telecom | τ²-Bench Telecom | 96.2 | 95 | 90.6 | OpenRouter model benchmarks |
| AA-LCR | AA-LCR | 66.3 | 63 | 65 | OpenRouter model benchmarks |
| GDPval-AA | GDPval-AA | 40.3 | 34.3 | 18.5 | OpenRouter model benchmarks |
| CritPt | CritPt | 12.9 | 7.1 | 2.9 | OpenRouter model benchmarks |
| SciCode | SciCode | 50 | 44.9 | 38.9 | OpenRouter model benchmarks |
| Terminal-Bench Hard | Terminal-Bench Hard | 46.2 | 35.6 | 35.6 | OpenRouter model benchmarks |
| AA-Omniscience Accuracy | AA-Omniscience Accuracy | 43.3 | 37.2 | 33.5 | OpenRouter model benchmarks |
| AA-Omniscience Non-Hallucination Rate | AA-Omniscience Non-Hallucination Rate | 6 | 4.2 | 18.3 | OpenRouter model benchmarks |
| MMLU-Pro | Knowledge and reasoning | — | — | 85 | OpenEvals/leaderboard-data |
| SWE-bench Pro | Coding | — | — | 15.56 | OpenEvals/leaderboard-data |
| SWE-bench Verified | Coding | — | — | 70 | OpenEvals/leaderboard-data |
| Terminal-Bench | Agentic coding | — | — | 39.6 | OpenEvals/leaderboard-data |
| AIME 2026 | Math | — | — | 94.17 | OpenEvals/leaderboard-data |