DeepSeek
As of 2026-10-18, DeepSeek has 17 models on the meo leaderboard, 2 first-party-tested. Its highest-ranked model is DeepSeek: DeepSeek V4 Flash 0423 (#11, 81.8%), while DeepSeek: DeepSeek V4 Pro 0423 leads on Effective Value (π).
Official: www.deepseek.com Β· huggingface.co/deepseek-ai
| meo rank | Model | Accuracy | π index | $ / correct | Context | Weights |
|---|---|---|---|---|---|---|
| #11 | DeepSeek: DeepSeek V4 Flash 0423 | 81.8% | 0.43 | $0.0016 | 1.0M | Open |
| #14 | DeepSeek: DeepSeek V4 Pro 0423 | 80.3% | 0.53 | $0.0182 | 1.0M | Open |
| β | DeepSeek: DeepSeek V3 | β | β | β | 164K | Open |
| β | DeepSeek: DeepSeek V3 0324 | β | β | β | 164K | Open |
| β | DeepSeek: DeepSeek V3.1 | β | β | β | 164K | Open |
| β | DeepSeek: DeepSeek V3.1 Terminus | β | β | β | 164K | Open |
| β | DeepSeek: DeepSeek V3.2 | β | β | β | 164K | Open |
| β | DeepSeek: DeepSeek V3.2 Exp | β | β | β | 164K | Open |
| β | DeepSeek: DeepSeek V4 Flash 0731 | β | β | β | 1.0M | Open |
| β | DeepSeek: DeepSeek V4 Flash Vision Exp | β | β | β | 1.0M | Open |
| β | DeepSeek: DeepSeek V4 Pro 0813 | β | β | β | 1.0M | Open |
| β | DeepSeek: DeepSeek V4.1 Flash | β | β | β | 1.0M | Open |
| β | DeepSeek: DeepSeek V4.1 Flash (batch) | β | β | β | 1.0M | Open |
| β | DeepSeek: R1 | β | β | β | 64K | Open |
| β | DeepSeek: R1 0528 | β | β | β | 164K | Open |
| β | DeepSeek: R1 Distill Llama 70B | β | β | β | 128K | Open |
| β | DeepSeek: R1 Distill Qwen 32B | β | β | β | 128K | Open |
Frequently asked questions
How many AI models does DeepSeek have?
DeepSeek has 17 models on the meo leaderboard, of which 2 are first-party-tested. Its best-ranked model is DeepSeek: DeepSeek V4 Flash 0423 at #11 (81.8% accuracy).
What is DeepSeek's best AI model?
On the un-leaked meo benchmark, DeepSeek's top model is DeepSeek: DeepSeek V4 Flash 0423, ranked #11 overall with 81.8% accuracy, while DeepSeek: DeepSeek V4 Pro 0423 leads the lab on Effective Value (π).
Does DeepSeek release open-weights models?
Yes β 17 of DeepSeek's 17 listed models are open-weights (self-hostable).
Related: all AI labs Β· full leaderboard Β· rankings Β· methodology.