DeepSeek V4
#6deepseek · v4 · seit 2026-04-24 · 392× · zuletzt 30. Juni 2026
70
Momentum
DeepSeek V4 is a family of open-weight language models (Mixture-of-Experts architecture) released by DeepSeek on April 24, 2026 as a preview. It comprises two variants: V4-Pro (1.6 trillion total parameters, 49B active) and V4-Flash (284B total, 13B active). Both variants support a 1-million-token context window by default and are available on Hugging Face under the MIT license. The model remains a text-only language model without multimodal capabilities.
Historique du momentum
04.04.03.07.
Fonctionnalités
| Key Benchmark (%) | V4-Pro-Max (per DeepSeek official, vendor-reported): SWE-bench Verified 80.6%; LiveCodeBench 93.5; Codeforces Rating 3206; MMLU-Pro 87.5%; GPQA Diamond 90.1% |
| Context Window (Tokens) | 1,000,000 tokens (1M) – standard for both variants (V4-Pro & V4-Flash); max output: 384,000 tokens |
| License | MIT License (for V4-Pro and V4-Flash, open weights on Hugging Face) |
| Multimodality | No multimodal support – V4 is a pure language model (text only); multimodality reportedly deferred due to compute constraints |
| Platform | Web (chat.deepseek.com), mobile app, API (api.deepseek.com); open weights on Hugging Face (deepseek-ai/DeepSeek-V4-Pro & DeepSeek-V4-Flash); OpenAI- and Anthropic-compatible API access |
| Price per 1M Tokens | V4-Flash: $0.14 input / $0.28 output; V4-Pro: $0.435 input / $0.87 output (cache hit significantly cheaper) |
| Release Date | April 24, 2026 (preview launch) |
