DeepSeek
ChinaModel developerFirst-party API
Hangzhou lab famous for frontier-level open models at rock-bottom cost.
What they do
DeepSeek shocked the field in January 2025 when its R1 reasoning model matched far more expensive Western models at a fraction of the training cost, publishing the methods openly. The V3/V4 series continues that trajectory, with Flash and Pro tiers and open technical reports.
Models they develop
31 in this directory.
- DeepSeek Coder V2 Lite InstructDeepSeekOpen weights
- DeepSeek LLM 67B Chat (V1)DeepSeek67B parametersOpen weights
- DeepSeek R1 (Jan '25)DeepSeekOpenReasoning
- DeepSeek R1 0528 (May '25)DeepSeekOpenReasoning
- DeepSeek R1 0528 Qwen3 8BDeepSeekOpen8B parametersReasoning
- DeepSeek R1 Distill Llama 70BDeepSeekOpen70B parametersReasoning
- DeepSeek R1 Distill Llama 8BDeepSeekOpen8B parametersReasoning
- DeepSeek R1 Distill Qwen 1.5BDeepSeekOpen1.5B parametersReasoning
- DeepSeek R1 Distill Qwen 14BDeepSeekOpen14B parametersReasoning
- DeepSeek R1 Distill Qwen 32BDeepSeekOpen32B parametersReasoning
- DeepSeek V3 (Dec '24)DeepSeekOpen weights
- DeepSeek V3 0324DeepSeekOpen weights
- DeepSeek V3.1 (Non-reasoning)DeepSeekOpen weights
- DeepSeek V3.1 (Reasoning)DeepSeekOpen weights
- DeepSeek V3.1 Terminus (Non-reasoning)DeepSeekOpen weights
- DeepSeek V3.1 Terminus (Reasoning)DeepSeekOpen weights
- DeepSeek V3.2 (Non-reasoning)DeepSeekOpen weights
- DeepSeek V3.2 (Reasoning)DeepSeekOpen weights
- DeepSeek V3.2 Exp (Non-reasoning)DeepSeekOpen weights
- DeepSeek V3.2 Exp (Reasoning)DeepSeekOpen weights
- DeepSeek V3.2 SpecialeDeepSeekOpen weights
- DeepSeek V4 Flash (Non-reasoning)DeepSeekOpenFrontier
- DeepSeek V4 Flash (Reasoning)DeepSeekOpenFrontier
- DeepSeek V4 Flash 0731 (Reasoning)DeepSeekOpenFrontier
- DeepSeek V4 Pro (Non-reasoning)DeepSeekOpenFrontier
- DeepSeek V4 Pro (Reasoning)DeepSeekOpenFrontier
- DeepSeek-Coder-V2DeepSeekOpen weights
- DeepSeek-V2-ChatDeepSeekOpen weights
- DeepSeek-V2.5DeepSeekOpen weights
- DeepSeek-V2.5 (Dec '24)DeepSeekOpen weights
- R1 1776DeepSeekOpenReasoning