Best cell per row highlighted. Null means undisclosed — never counted as zero.
| Provider | DeepSeek | DeepSeek | DeepSeek |
| Family / tier | DeepSeek V3 · Large | DeepSeek V4 · Flash | DeepSeek V4 · Pro |
| Status | GA | GA | GA |
| Released | 2025-12-01 | 2026-09-10 | 2026-08-13 |
| Knowledge cutoff | 2025-07 | — | 2026-02 |
| Context window | 128K | 1M | 1M |
| Max output tokens | 64K | 384K | 384K |
| Input modalities | text | text, image | text |
| Reasoning mode | optional | optional | optional |
| Open weights | Yes | Yes | Yes |
| Input | $0.252 | $0.3 | $1.32 |
| Output | $0.378 | $1.2 | $3.96 |
| Cached input | $0.025 | $0.006 | $0.044 |
| Batch input | — | — | — |
| Free tier | No | Yes | Yes |
| Output speed (tok/s) | — | — | — |
| Time to first token (s) | — | — | — |
| Latency tier | medium | fast | medium |
| Trains on inputs | Yes | Yes | Yes |
| License | MIT | MIT | MIT |
| Clouds | — | — | — |
| Compliance | — | — | — |
| Unified score | 8.2 | 8.6 | 8.5 |
| Decision Maker | 8 | 8.5 | 8 |
| Domain Strategist | 8 | 9 | 8.5 |
| Finance Lead | 9 | 9 | 9 |
| Domain Practitioner | 8.5 | 9 | 8.5 |
| Power User | 7.5 | 8.5 | 7.5 |
| Skeptic | 7.5 | 7.5 | 7.5 |
| Value score | 9.3 | 9.6 | 9.4 |
| MMLU-Pro | 85 | — | 87.5 |
| GPQA Diamond | — | 90.9 | 90.1 |
| AIME 2025 | 93.1 | — | — |
| SWE-bench Verified | 67.8 | — | 80.6 |
| HumanEval | — | — | 76.8 |
| LiveCodeBench | 74.1 | — | 93.5 |
| Terminal-Bench | — | 90.6 | — |
| MRCR Long Context | — | — | 83.5 |
| SimpleQA | — | — | 57.9 |
| Humanity's Last Exam | 30.6 | 36.8 | 37.7 |
| LMArena Elo | 1425 | — | 1463 |
| LMArena Coding Elo | 1476 | — | 1503 |
Scores link to their sources. Missing cells mean the vendor hasn't disclosed a result — honesty over padding.