- Size
- 291B parameterscounted from the safetensors header on the Hub
- Licence
- unknown, MIT LicenseOur reading, check the licence: You can use, change and redistribute the model for any purpose, including commercially. Keep the copyright notice and the licence text with copies. You do not have to share your changes.
- On the Hub since
- 22 Apr 2026the date the repo was created
- Context length
- 1M tokensfrom the model's config
- Kind
- Base model (no base model declared on the Hub)DeepseekV4ForCausalLM
- On Hugging Face
- deepseek-ai/DeepSeek-V4-Flash1.2M downloads in the last 30 days, 2.4k likes
Scores
- effort max7.1%via Epoch AI
- effort high3.4%via Epoch AI
Epoch's name: deepseek-v4-flash. Best closed model on this set: gemini-4-carbon_high, 36.0% (Google DeepMind). All 271 results
- effort max86.4%conceptualreasoning.ai
Epoch's name: deepseek-v4-flash. Best closed model on this set: claude-fable-5-2_unknown, 99.2% (Anthropic). All 241 results
- effort max35.9%31.4 to 40.4conceptualreasoning.ai
Epoch's name: deepseek-v4-flash. Best closed model on this set: claude-fable-5-2_xhigh, 70.6% (Anthropic). All 203 results
- effort unknown46.3%lmcouncil.ai
Epoch's name: deepseek-v4-flash. Best closed model on this set: claude-fable-5-2_unknown, 94.9% (Anthropic). All 125 results
- effort max45.6%±0.0htihle.github.io
- effort high43.8%±0.0htihle.github.io
Epoch's name: deepseek-v4-flash. Best closed model on this set: gpt-6-astra_promax, 93.6% (OpenAI). All 173 results
Claimed by the maker Claimed
The model card has no claimed scores in a machine-readable form (the card's model-index), so none are shown. The maker may give scores in the card text: read the card.
Other views, which we link to and do not copy: look this model up on Artificial Analysis and the Arena leaderboard.
Independent scores from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Each row links its own source. What the badges mean.
Run it yourself
Memory needed = weights + context cache + 0.50 GB for the runtime. Context cache at 8k tokens: 0.67 GB, worked out from the model's config (deepseek-ai/DeepSeek-V4-Flash), at 16-bit precision. Choose a machine above to see a verdict for each format. Method
GGUF for llama.cpp, Ollama, LM Studio and others
unsloth/DeepSeek-V4-Flash-GGUF (converted by unsloth), 19k downloads. Sizes are the real files on the Hub, all parts added up.
| Format | Size | Memory needed | Get it |
|---|---|---|---|
| UD-IQ3_XXS | 95.9 GBfile size, 4 parts | 97.1 GB95.9 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-IQ3_XXSDownload hf download unsloth/DeepSeek-V4-Flash-GGUF --include "UD-IQ3_XXS/DeepSeek-V4-Flash-UD-IQ3_XXS-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-GGUF |
| UD-IQ3_S | 109.3 GBfile size, 4 parts | 110.4 GB109.3 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-IQ3_SDownload hf download unsloth/DeepSeek-V4-Flash-GGUF --include "UD-IQ3_S/DeepSeek-V4-Flash-UD-IQ3_S-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-GGUF |
| UD-Q3_K_M | 120.4 GBfile size, 4 parts | 121.6 GB120.4 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-Q3_K_MDownload hf download unsloth/DeepSeek-V4-Flash-GGUF --include "UD-Q3_K_M/DeepSeek-V4-Flash-UD-Q3_K_M-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-GGUF |
| UD-Q3_K_XL | 120.6 GBfile size, 4 parts | 121.7 GB120.6 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-Q3_K_XLDownload hf download unsloth/DeepSeek-V4-Flash-GGUF --include "UD-Q3_K_XL/DeepSeek-V4-Flash-UD-Q3_K_XL-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-GGUF |
| UD-IQ4_NL | 128.4 GBfile size, 4 parts | 129.6 GB128.4 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-IQ4_NLDownload hf download unsloth/DeepSeek-V4-Flash-GGUF --include "UD-IQ4_NL/DeepSeek-V4-Flash-UD-IQ4_NL-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-GGUF |
| UD-IQ4_XS | 128.4 GBfile size, 4 parts | 129.6 GB128.4 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-IQ4_XSDownload hf download unsloth/DeepSeek-V4-Flash-GGUF --include "UD-IQ4_XS/DeepSeek-V4-Flash-UD-IQ4_XS-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-GGUF |
| UD-Q4_K_XL | 144.4 GBfile size, 5 parts | 145.6 GB144.4 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-Q4_K_XLDownload hf download unsloth/DeepSeek-V4-Flash-GGUF --include "UD-Q4_K_XL/DeepSeek-V4-Flash-UD-Q4_K_XL-*-of-00005.gguf" --local-dir DeepSeek-V4-Flash-GGUF |
| UD-Q8_K_XL | 150.8 GBfile size, 5 parts | 151.9 GB150.8 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-Q8_K_XLDownload hf download unsloth/DeepSeek-V4-Flash-GGUF --include "UD-Q8_K_XL/DeepSeek-V4-Flash-UD-Q8_K_XL-*-of-00005.gguf" --local-dir DeepSeek-V4-Flash-GGUF |
MLX for Apple silicon
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| MLXmlx-community/DeepSeek-V4-Flash | 144.4 GBfile size | 145.6 GB144.4 + 0.67 + 0.50 | Download page |
| MLX 8bitmlx-community/deepseek-ai-DeepSeek-V4-Flash-8bit | 281.5 GBfile size | 282.7 GB281.5 + 0.67 + 0.50 | Download page |
Other conversions (14; sizes not read)
- nvidia/DeepSeek-V4-Flash-NVFP4 NVFP4, 111k downloads
- sgl-project/DeepSeek-V4-Flash-FP8 FP8, 22k downloads
- RedHatAI/DeepSeek-V4-Flash-NVFP4-FP8 NVFP4, 939 downloads
- unsloth/DeepSeek-V4-Flash FP8, 474 downloads
- nvidia/DeepSeek-V4-Flash-nvfp4-DSpark NVFP4, 474 downloads
- intel-ai/DeepSeek-V4-Flash-MXFP4-Mixed-CT-AutoRound Other, mxfp4, 468 downloads
- amd/DeepSeek-V4-Flash-MXFP4 Other, mxfp4, 347 downloads
- Pinaster/DeepSeek-V4-Flash-FP8-4layer FP8, 265 downloads
- canada-quant/DeepSeek-V4-Flash-W4A16-FP8-MTP GPTQ, w4a16, 211 downloads
- 0xSero/DeepSeek-V4-Flash-180B FP8, 204 downloads
- Intel/DeepSeek-V4-Flash-W4A16-AutoRound Other, w4a16, 166 downloads
- Entrit/DeepSeek-V4-Flash-d2swap-mp2 Other, 162 downloads
- amd/DeepSeek-V4-Flash-NVFP4 NVFP4, 135 downloads
- canada-quant/DeepSeek-V4-Flash-W4A16-FP8 GPTQ, w4a16, 122 downloads
The Hub tags 100 or more quantised repos as conversions of this model; we read the most downloaded and show the maker's own first.
Run it by API
9 providers serve this model through the Hugging Face router. Prices are US dollars per million tokens. Precision is what the provider reports; a lower precision than the original weights can change answers. Live first, then cheapest output.
| Provider | In | Out | Context | Precision | Tokens/s | First token | Uptime 24h | Tools | JSON |
|---|---|---|---|---|---|---|---|---|---|
| Wafer | $0.12 | $0.17 | 1M | mxfp4 | 61 | 0.4 s | 100.0% | yes | yes |
| DeepInfra | $0.09 | $0.18 | 1M | fp8 | 78 | 0.6 s | 100.0% | yes | yes |
| Venice | $0.097 | $0.19 | 977k | not given | 83 | 0.9 s | 100.0% | yes | yes |
| Parasail | $0.14 | $0.28 | 1M | fp8 | 123 | 0.5 s | 99.7% | yes | yes |
| SiliconFlow | $0.13 | $0.28 | 1M | fp8 | 56 | 1.1 s | 98.6% | yes | no |
| Relace | $0.05 | $0.45 | 1M | fp4 | 146 | 0.5 s | 99.9% | yes | no |
| Mancer 2 | $0.15 | $0.50 | 1M | fp8 | 70 | 0.9 s | 100.0% | yes | yes |
| Azure | $0.21 | $0.56 | 1M | not given | 95 | 1.1 s | 99.5% | yes | no |
| featherless-ai | ? | ? | ? | not given | ? | ? | ? | not given | not given |
Through the router, with a Hugging Face token (the router bills your Hugging Face account). Change the name after the colon to pick another provider.
curl https://router.huggingface.co/v1/chat/completions \
-H "Authorization: Bearer $HF_TOKEN" -H "Content-Type: application/json" \
-d '{"model": "deepseek-ai/DeepSeek-V4-Flash:wafer", "messages": [{"role": "user", "content": "Hello"}]}'Not built yet: one key for every open model, with BlockRun, and our own tested-quality score per provider. Until then, precision and measured speed are the best guide to a provider here.
Fine-tunes of this model
Repos that declare this model as their base, most downloaded first. Listing is not a recommendation; unreviewed ones have not been checked.
- MemTensor/MemT-1.0-Flash288B58kUnreviewed
Planned for this page Not built yet
- Try it in private chat
- A link to Solana AI's confidential chat for the models it serves. This model is not listed as served there.
- Fine-tune it
- Fine-tuning a few base models from this page and publishing the result here.
- Creators and earnings
- Pages for people who publish fine-tunes and adapters of this model, and what they earn.
- Fund and bounties
- Bounties for new open models, paid from a public fund. How the fund will work.
- Contribute compute
- How to give compute to open model work, with Akash.
Model data from the Hugging Face Hub 19 Nov 2026, 13:43 UTC. The repo. Found a mistake on this page? Tell us.