Skip to content
NEWLive updates feed from Hugging Face, arXiv and GitHub  See what changed →

deepseek-ai · deepseek_v4

DeepSeek-V4-Flash-0731

DeepSeek-V4-Flash-0731 by deepseek-ai: a 304.2B-parameter mixture-of-experts open-weight model under the mit license.

mitMixture of experts1049K contextAuto-imported · not yet reviewed

4,178,593 downloads · 3,980 likes · Model card · synced 2026-09-19

Specifications

Parameters
304.2B (7.1B active)
Architecture
moe · deepseek_v4
Layers
43
Hidden size
4,096
KV heads · head dim
1 · 512
Vocabulary
129,280
Native dtype
bfloat16
Context window
1,048,576 tokens
Released
2026-07-31
Last modified on Hub
2026-08-01

Features & licensing

License
mit
Commercial use
Yes
Modalities
text
Tool / function calling
No
Reasoning mode
No
Languages
EN
Quantised variants on Hub
FP8, GGUF, bitsandbytes
Pipeline
text-generation
transformerssafetensorsdeepseek_v4text-generationeval-resultsendpoints_compatible8-bitfp8

Task fit

Editorial scores (0–100) used by the recommendation engine.

  • Chat70
  • RAG / Q&A70
  • Code70
  • Summarisation70
  • Extraction70
  • Agents70

Hardware to run DeepSeek-V4-Flash-0731

Weights need about 669 GB at FP16, 351 GB at INT8 and 201 GB at INT4. The KV cache adds roughly 88 MB per 1,000 tokens per request. For a reference workload of 5 requests per second with 1,500 input and 300 output tokens, total GPU memory is around 269.6 GB, which fits on 4 × NVIDIA A100 (80 GB).

GPU configurations at the reference workload
ConfigurationVRAMUtilisationEst. first tokenCloud / month
7 × NVIDIA L40S (48 GB)336 GB80 %~0.2 s$9,709
4 × NVIDIA A100 (80 GB)recommended320 GB84 %~0.1 s$9,344
4 × NVIDIA H100 (80 GB)320 GB84 %~0.0 s$13,140

Estimates only. Run an assessment for your own traffic.

Auto-imported from Hugging Face. Architecture, license and feature data are synced from the model repository; the task-fit scores below are provisional defaults until a curator reviews them.