DeepSeek
9 ranked models in a catalog of 15 text models. DeepSeek V4.1 Flash leads this lab at #31 overall.
Ranked models
Global ranks · current snapshot| Rank | Model | SI Score | coding | math | reasoning | preference | Confidence | Blended price | Context | Released |
|---|---|---|---|---|---|---|---|---|---|---|
| #31 | DeepSeek V4.1 FlashOpen weights | 66.2 | 73.3 | 88.0 | 82.7 | 78.8 | 69% | $0.26 | 1M | Sep 10, 2026 |
| #44 | DeepSeek V4 ProOpen weights | 63.1 | 58.1 | 67.5 | 80.3 | 77.8 | 85% | $1.98 | 1M | Apr 24, 2026 |
| #46 | DeepSeek V4 Pro 0813Open weights | 63.0 | 65.4 | 77.7 | 82.0 | — | 69% | $0.99 | 1M | Aug 12, 2026 |
| #60 | DeepSeek V4 Flash 0731Open weights | 61.0 | 59.7 | 71.3 | 82.9 | — | 69% | — | 1M | Jul 31, 2026 |
| #79 | DeepSeek V4 FlashOpen weights | 56.2 | 52.4 | 74.5 | 69.5 | 76.1 | 53% | $0.52 | 1M | Apr 24, 2026 |
| #82 | DeepSeek V3 0324Open weights | 55.7 | 48.5 | 63.0 | 67.6 | 70.5 | 74% | — | 164K | Mar 24, 2025 |
| #93 | DeepSeek-V3Open weights | 54.1 | — | 48.5 | 56.5 | 66.0 | 93% | $0.48 | 131K | Dec 26, 2024 |
| #96 | DeepSeek V3.2Open weights | 53.3 | 70.0 | — | 31.6 | 75.0 | 56% | $0.31 | 128K | Dec 1, 2025 |
| #113 | DeepSeek-R1Open weights | 47.9 | 44.5 | 86.6 | 18.9 | 70.3 | 88% | $0.96 | 128K | Jan 20, 2025 |
Pillars use a 0–100 scale. * marks an imputed neutral prior where the model has no scored results in that pillar; it is not a measured benchmark result. Prices are USD per 1M tokens; see the blend and price comparison.
Scores and releases
Provisional models
6 awaiting rank-eligible evidence- DeepSeek V4 Flash Vision Exp
Evidence completeness 21.0%; at least 50% required
- DeepSeek OCR 2
No usable open benchmark results found
- DeepSeek Chat
Evidence completeness 32.0%; at least 50% required
- DeepSeek Reasoner
Evidence completeness 32.0%; at least 50% required
- DeepSeek-V3.1
Results cover 1 pillar; at least 2 required
- DeepSeek-R1-Distill-Qwen-32B
Evidence completeness 32.0%; at least 50% required