Model vendor
Chinese AI company developing open and frontier language models, including the DeepSeek-V3 and DeepSeek-R1 model families.
Updated Aug 10, 2026. Data as of 2026-08-07
Default-version specifications and scored-version capability values are labeled separately.
| DE | DeepSeek V4 Pro | 80.1 | 100.0% | 1M | Open weights | DeepSeek-V4-Pro-Max | |
| DE | DeepSeek-V3.2-Speciale | 75.5 | 60.0% | 131.1K | Closed weights | DeepSeek-V3.2-Speciale | |
| DE | DeepSeek-V4-Flash-0731 | 71.7 | 100.0% | 1M | Open weights | DeepSeek-V4-Flash-Max | |
| DE | DeepSeek-V2.5 | 69.5 | 60.0% | 8.2K | Closed weights | DeepSeek-V2.5 | |
| DE | DeepSeek-R1-0528 | 44.1 | 100.0% | 131.1K | Closed weights | DeepSeek-R1-0528 | |
| DE | DeepSeek-V3 0324 | 38.1 | 80.0% | 163.8K | Closed weights | DeepSeek-V3 0324 | |
| DE | DeepSeek-V3.1 | 36.0 | 100.0% | 131.1K | Open weights | DeepSeek-V3.1 | |
| DE | DeepSeek R1 Distill Llama 8B | N/A | N/A | N/A | Closed weights | Not scored | |
| DE | DeepSeek R1 Distill Qwen 7B | N/A | N/A | N/A | Closed weights | Not scored | |
| DE | DeepSeek VL2 Tiny | N/A | N/A | N/A | Closed weights | Not scored |
Each unique model uses its latest scored version.
Leaders from this vendor's current published catalog.
Vendor context from the published model registry.
Chinese AI company developing open and frontier language models, including the DeepSeek-V3 and DeepSeek-R1 model families.
Model ownership is separate from API hosting. One model can be offered by multiple API providers with different prices.
Vendor description and inventory come from the published model registry.
Common questions about DeepSeek.
DeepSeek-V4-Pro currently has the strongest published LLMBoard score from this vendor.
The current public catalog contains 10 independently selectable DeepSeek model products and 32 published versions.
Not necessarily. Vendor identifies the model developer; API providers are tracked separately.
3 current default versions are marked as open weight.