DeepSeek-V4.1-Flash (2026-09-10)
- Open weights
Prices
0 offerings · prices as listedNo priced offering is listed for this model yet.
Details
- Family
- deepseek-flash
Run locally
Weights: deepseek-ai/DeepSeek-V4.1-Flash- Parameters 763.2B
- Active 228.1B (estimated) · 384 experts, 6 per token
- Native precision FP8
- Layers 40
- KV cache per token 80 KB
- Config positions 1024K (config.json max_position_embeddings)
- Needs
- ≈ 397.1 GB (weights 377.6 + KV cache 0.6 + overhead 18.9)
- Usable
- 21.6 GB (90% of its memory)
- Try
- it’s too large for this hardware; see the smallest that runs it below
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| FP8 (native)maker’s file | 475.3 GB | ≈ 499.6 | ✕Doesn’t fit | — |
| Q6_K | 583.7 GB | ≈ 613.5 | ✕Doesn’t fit | — |
| Q5_K_M | 505.5 GB | ≈ 531.5 | ✕Doesn’t fit | — |
| Q5_0 | 488.7 GB | ≈ 513.7 | ✕Doesn’t fit | — |
| Q4_K_M | 431.8 GB | ≈ 454.0 | ✕Doesn’t fit | — |
| Q4_0 | 399.8 GB | ≈ 420.4 | ✕Doesn’t fit | — |
| IQ4_XS | 377.6 GB | ≈ 397.1 | ✕Doesn’t fit | — |
| Q3_K_Mquality drops | 347.4 GB | ≈ 365.4 | ✕Doesn’t fit | — |
| IQ3_XXSquality drops | 271.9 GB | ≈ 286.1 | ✕Doesn’t fit | — |
| Q2_Kquality drops | 266.5 GB | ≈ 280.5 | ✕Doesn’t fit | — |
| IQ2_XXSquality drops | 183.0 GB | ≈ 192.8 | ✕Doesn’t fit | — |
Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.
Compared with other hardware
Each at the highest precision it runs. Click one for details.
Smallest that runs it (Q4_K_M, 8K): no single listed machine; about 7 × 80 GB GPUs.
Run commands
The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.
Memory to fine-tune
Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.
Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated
Cited scores
From third-party leaderboards, not measured by us- LMArena Text 1,476
#28 of 163 listed models · max effort · CI 1,470–1,482 · 13,657 votes
- LMArena Text · Chinese 1,523
#25 of 161 listed models · max effort · CI 1,504–1,542 · 1,032 votes
- LMArena Text · Coding 1,523
#25 of 163 listed models · max effort · CI 1,513–1,533 · 3,745 votes
- LMArena Text · Creative writing 1,441
#39 of 163 listed models · max effort · CI 1,429–1,453 · 2,763 votes
- LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Oct 19 – 28, 2026
- changes we made
News
All news →- Sep 10, 2026Release DeepSeek releases DeepSeek-V4.1-Flash (2026-09-10) DeepSeek-V4.1-Flash (2026-09-10) is now available.
Changes
All changes →- New language model
Sources
- deepseek-ai/DeepSeek-V4.1-Flash model card (opens in a new tab) huggingface.co · checked Nov 1, 2026
- DeepSeek API change log (opens in a new tab) api-docs.deepseek.com · checked Nov 2, 2026
- deepseek-ai/DeepSeek-V4.1-Flash config.json (opens in a new tab) huggingface.co · checked Nov 3, 2026
- deepseek-ai/DeepSeek-V4.1-Flash files (opens in a new tab) huggingface.co · checked Nov 5, 2026
DeepSeek’s official policies: data use, privacy, terms →
Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data
DeepSeek · China · deepseek.com
Related models
Snapshots, the same family and other models from this provider