71Fit score
Legacy compatibility id mapping to DeepSeek V4 Flash non-thinking mode (per DeepSeek pricing notes).
Model details →Open-weight models you can run on your own infrastructure. No vendor lock-in, no per-token cost.
Legacy compatibility id mapping to DeepSeek V4 Flash non-thinking mode (per DeepSeek pricing notes).
Model details →Legacy compatibility id mapping to DeepSeek V4 Flash thinking mode (per DeepSeek pricing notes).
Model details →DeepSeek V4 Flash — 1M context, thinking and non-thinking modes; see DeepSeek pricing/docs for current capabilities.
Model details →| # | Model | Provider | Context | Input price / 1M | Tier |
|---|---|---|---|---|---|
| 1 | DeepSeek Chat (legacy id) | DeepSeek | 1M Ultra context (1M+) | Custom | Variable |
| 2 | DeepSeek Reasoner (legacy id) | DeepSeek | 1M Ultra context (1M+) | Custom | Variable |
| 3 | DeepSeek V4 Flash | DeepSeek | 1M Ultra context (1M+) | Custom | Variable |
| 4 | DeepSeek V4 Pro | DeepSeek | 1M Ultra context (1M+) | Custom | Variable |
| 5 | DeepSeek R1 | DeepSeek | 128K Long context (128K+) | Custom | Variable |
| 6 | DeepSeek V3 | DeepSeek | 128K Long context (128K+) | Custom | Variable |
| 7 | Qwen2.5 72B Instruct | Qwen | 128K Long context (128K+) | Custom | Variable |
| 8 | Qwen2.5 Coder 32B | Qwen | 128K Long context (128K+) | Custom | Variable |
| 9 | o1 | OpenAI | 200K Long context (128K+) | Custom | Variable |
| 10 | o3-mini | OpenAI | 200K Long context (128K+) | Custom | Variable |
| 11 | GPT-4o | OpenAI | 128K Long context (128K+) | Custom | Variable |
| 12 | GPT-4o mini | OpenAI | 128K Long context (128K+) | Custom | Variable |
| 13 | Llama 3.1 405B Instruct | Meta | 128K Long context (128K+) | Custom | Variable |
| 14 | Llama 3.1 70B Instruct | Meta | 128K Long context (128K+) | Custom | Variable |
| 15 | Llama 3.1 8B Instruct | Meta | 128K Long context (128K+) | Custom | Variable |
The trained model weights are publicly downloadable and you can run them on your own hardware. Different from 'open source' which would also include training code and data — most open-weight models are not fully open source.