- Size
- 304B 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
- 31 Jul 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-Flash-07311.8M downloads in the last 30 days, 4.1k likes
Scores
- effort max32.0%±4.713 Oct 2026, log
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-fable-5-2_max, 79.0% (Anthropic). All 166 results
- effort max24.4%±6.82 Aug 2026, log
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-fable-5-2_max, 100.0% (Anthropic). All 83 results
- effort max57.5%±2.92 Aug 2026, log
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-fable-5-2_max, 95.1% (Anthropic). All 127 results
- effort max91.0%±1.82 Aug 2026, log
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: gpt-6.2-astra_max, 96.8% (OpenAI). All 341 results
- effort max34.0%±4.85 Aug 2026, Epoch AI
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: gpt-6.2-astra_max, 95.0% (OpenAI). All 148 results
- effort max94.4%±2.12 Aug 2026, log
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-fable-5_high, 100.0% (Anthropic). All 318 results
- effort max33.6%±1.527 Aug 2026, log
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-fable-5-2_max, 94.0% (Anthropic). All 105 results
- effort max89.0%arcprize.org
- effort high87.0%arcprize.org
- effort low84.0%arcprize.org
- effort none11.8%arcprize.org
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-fable-5_max, 98.5% (Anthropic). All 318 results
- effort max61.4%via Epoch AI
- effort high56.0%via Epoch AI
- effort low46.0%via Epoch AI
- effort none2.1%via Epoch AI
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: gpt-6-astra_max, 95.0% (OpenAI). All 299 results
- effort max16.6%via Epoch AI
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: gemini-4-carbon_high, 36.0% (Google DeepMind). All 271 results
- effort max90.9%conceptualreasoning.ai
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-fable-5-2_unknown, 99.2% (Anthropic). All 241 results
- effort unknown; Harness: devin-cli; Reasoning effort: high18.8%cognition.com
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-opus-5-5_medium, 54.6% (Anthropic). All 46 results
- effort max41.7%37.5 to 46.0conceptualreasoning.ai
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-fable-5-2_xhigh, 70.6% (Anthropic). All 203 results
- effort unknown56.0%±5.0vals.ai
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-fable-5-1_max, 100.0% (Anthropic). All 85 results
- effort unknown61.1%lmcouncil.ai
Epoch's name: deepseek-v4-flash-0731. Best closed model on this set: claude-fable-5-2_unknown, 94.9% (Anthropic). All 125 results
- effort max63.0%±0.0htihle.github.io
- effort high57.0%±0.0htihle.github.io
Epoch's name: deepseek-v4-flash-0731. 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-0731), 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-0731-GGUF (converted by unsloth), 112k downloads. Sizes are the real files on the Hub, all parts added up.
| Format | Size | Memory needed | Get it |
|---|---|---|---|
| UD-IQ3_XXS | 97.1 GBfile size, 4 parts | 98.2 GB97.1 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-IQ3_XXSDownload hf download unsloth/DeepSeek-V4-Flash-0731-GGUF --include "UD-IQ3_XXS/DeepSeek-V4-Flash-0731-UD-IQ3_XXS-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-0731-GGUF |
| UD-IQ3_S | 108.1 GBfile size, 4 parts | 109.3 GB108.1 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-IQ3_SDownload hf download unsloth/DeepSeek-V4-Flash-0731-GGUF --include "UD-IQ3_S/DeepSeek-V4-Flash-0731-UD-IQ3_S-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-0731-GGUF |
| UD-Q3_K_M | 119.3 GBfile size, 4 parts | 120.5 GB119.3 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q3_K_MDownload hf download unsloth/DeepSeek-V4-Flash-0731-GGUF --include "UD-Q3_K_M/DeepSeek-V4-Flash-0731-UD-Q3_K_M-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-0731-GGUF |
| UD-Q3_K_XL | 119.4 GBfile size, 4 parts | 120.6 GB119.4 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-Q3_K_XLDownload hf download unsloth/DeepSeek-V4-Flash-0731-GGUF --include "UD-Q3_K_XL/DeepSeek-V4-Flash-0731-UD-Q3_K_XL-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-0731-GGUF |
| UD-IQ4_NL | 127.3 GBfile size, 4 parts | 128.4 GB127.3 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-IQ4_NLDownload hf download unsloth/DeepSeek-V4-Flash-0731-GGUF --include "UD-IQ4_NL/DeepSeek-V4-Flash-0731-UD-IQ4_NL-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-0731-GGUF |
| UD-IQ4_XS | 127.3 GBfile size, 4 parts | 128.4 GB127.3 + 0.67 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/DeepSeek-V4-Flash-0731-GGUF:UD-IQ4_XSDownload hf download unsloth/DeepSeek-V4-Flash-0731-GGUF --include "UD-IQ4_XS/DeepSeek-V4-Flash-0731-UD-IQ4_XS-*-of-00004.gguf" --local-dir DeepSeek-V4-Flash-0731-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-0731-GGUF:UD-Q4_K_XLDownload hf download unsloth/DeepSeek-V4-Flash-0731-GGUF --include "UD-Q4_K_XL/DeepSeek-V4-Flash-0731-UD-Q4_K_XL-*-of-00005.gguf" --local-dir DeepSeek-V4-Flash-0731-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-0731-GGUF:UD-Q8_K_XLDownload hf download unsloth/DeepSeek-V4-Flash-0731-GGUF --include "UD-Q8_K_XL/DeepSeek-V4-Flash-0731-UD-Q8_K_XL-*-of-00005.gguf" --local-dir DeepSeek-V4-Flash-0731-GGUF |
MLX for Apple silicon
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| MLX 4bitmlx-community/DeepSeek-V4-Flash-0731-2.4bit-mixed | 86.5 GBfile size | 87.6 GB86.5 + 0.67 + 0.50 | Download page |
| MLXJundot/DeepSeek-V4-Flash-0731-oQ4e-mtp | 154.4 GBfile size | 155.6 GB154.4 + 0.67 + 0.50 | Download page |
Other conversions (15; sizes not read)
- 0xSero/deepseek-v4-flash-0731-spark EXL3, 330k downloads
- AtlasCloud/DeepSeek-V4-Flash-0731-FP8-DSpark FP8, 5.8k downloads
- nvidia/DeepSeek-V4-Flash-0731-NVFP4 NVFP4, 2.1k downloads
- intel-ai/DeepSeek-V4-Flash-0731-MXFP4-Mixed-CT-AutoRound Other, mxfp4, 1.4k downloads
- RedHatAI/DeepSeek-V4-Flash-0731-NVFP4 NVFP4, 1.1k downloads
- unsloth/DeepSeek-V4-Flash-0731 FP8, 623 downloads
- 0xSero/DeepSeek-V4-Flash-0731-REAP FP8, 526 downloads
- MJPansa/DeepSeek-V4-Flash-0731-NVFP4 NVFP4, 377 downloads
- Akan4ik/DeepSeek-V4-Flash-0731-FFT-BF16-Cyber FP8, bf16, 321 downloads
- Avifenesh/DeepSeek-V4-Flash-0731-NVFP4 NVFP4, 239 downloads
- turboderp/DeepSeek-V4-Flash-0731-exl3 EXL3, 227 downloads
- sokada4/DeepSeek-V4-Flash-0731-GPTQ-Int4-V1 GPTQ, int4, 168 downloads
- Jon-Nielsen/DeepSeek-V4-Flash-0731-MXFP4-INT8 Other, mxfp4, 162 downloads
- Laplace1313/DeepSeek-V4-Flash-0731-JA-REAP-K216-EXL3-3bpw-DGX-Spark EXL3, 142 downloads
- anoane/DeepSeek-V4-Flash-0731-exl3-2.54bpw EXL3, 103 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
19 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 |
|---|---|---|---|---|---|---|---|---|---|
| Relace | $0.07 | $0.14 | 1M | fp4 | 74 | 0.8 s | 99.9% | yes | no |
| DeepInfra | $0.06 | $0.18 | 1M | fp8 | 43 | 0.7 s | 99.0% | yes | yes |
| DigitalOcean | $0.12 | $0.24 | 1M | not given | 121 | 0.9 s | 100.0% | yes | yes |
| BaseTen | $0.13 | $0.26 | 1M | fp8 | 118 | 0.4 s | 100.0% | yes | no |
| Venice | $0.13 | $0.26 | 977k | not given | 60 | 0.7 s | 99.9% | yes | yes |
| CoreWeave | $0.13 | $0.27 | 256k | fp8 | 85 | 0.6 s | 100.0% | yes | yes |
| Cohere | $0.14 | $0.28 | 1M | fp8 | 116 | 0.4 s | 100.0% | yes | yes |
| Nebius | $0.14 | $0.28 | 1M | fp8 | 92 | 1.2 s | 100.0% | yes | yes |
| Parasail | $0.14 | $0.28 | 1M | fp8 | 27 | 0.8 s | 99.9% | yes | yes |
| Sail Research | $0.15 | $0.30 | 1M | fp4 | 33 | 1.2 s | 99.7% | yes | yes |
| Wafer | $0.062 | $0.30 | 1M | nvfp4 | 60 | 0.4 s | 100.0% | yes | yes |
| Reka | $0.10 | $0.42 | 256k | not given | 106 | 0.8 s | 100.0% | yes | yes |
| Inceptron | $0.13 | $0.49 | 1M | fp4 | 52 | 0.9 s | 100.0% | yes | yes |
| Mancer 2 | $0.15 | $0.50 | 1M | fp8 | 45 | 1.2 s | 98.0% | yes | yes |
| SiliconFlow | $0.22 | $0.66 | 1M | fp8 | 37 | 1.1 s | 99.3% | yes | yes |
| Novita | $0.41 | $1.23 | 1M | fp8 | 59 | 1.0 s | 100.0% | yes | yes |
| Cloudflare | $0.44 | $1.32 | 1M | not given | 68 | 1.3 s | 100.0% | yes | yes |
| PrimeIntellect | $0.44 | $1.32 | 1M | fp4 | 221 | 0.4 s | 100.0% | yes | yes |
| 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-0731:relace", "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.
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.