V4-Pro
#30deepseek · v4 · seit 24. April 2026 (Preview) · 26× · zuletzt 30. Juni 2026
37
Momentum
DeepSeek V4-Pro is the flagship model of the DeepSeek-V4 series (preview release), a Mixture-of-Experts model with 1.6 trillion total parameters and 49 billion active parameters per token. It supports a 1-million-token context window and is available as an open-weight model under the MIT license on Hugging Face. The model uses a new hybrid attention architecture (Compressed Sparse Attention + Heavily Compressed Attention) that requires only 27% of inference FLOPs and 10% of the KV cache of DeepSeek-V3.2 at 1M-token context. Three configurable reasoning modes (Non-Think, Think High, Think Max) allow trade-offs between latency and reasoning depth.
Historique du momentum
04.04.03.07.
Fonctionnalités
| Key Benchmark (%) | SWE-bench Verified: 80.6% | LiveCodeBench: 93.5% | GPQA Diamond: 90.1% | Codeforces: 3206 Rating (each V4-Pro-Max, vendor-reported) |
| Context Window (Tokens) | 1,048,576 tokens (1M); max. output: 384,000 tokens |
| License | MIT License (Open Weight) |
| Multimodality | Text only (no image input in current preview release) |
| Platform | DeepSeek API (deepseek-v4-pro), chat.deepseek.com (Expert Mode), open weights on Hugging Face (deepseek-ai/DeepSeek-V4-Pro); compatible with OpenAI ChatCompletions & Anthropic API |
| Price per 1M Tokens | Input (cache miss): $0.435 | Input (cache hit): $0.003625 | Output: $0.87 (after permanent 75% discount; list price: $1.74/$3.48) |
| Release Date | April 24, 2026 (preview release) |
