Ctrl K
Back to models

DeepSeek

DeepSeek-V4-Flash-0731

  • Released
  • Open weights
  • MIT
Compare
Intelligence Index

Max setting

34.3
Measured · Artificial Analysis v4.3.2
API price from
$0.44 in / $1.32 out / 1M tokOpenRouter · Cloudflare
Context window
1.05M tok

Performance profile

ReasoningArchived source
90.8%
GPQA Diamond
CodingArchived source
50.3%
SciCode
Cost efficiencyArchived source
156.3pts / $
Intelligence ÷ eval cost

Capabilities

Context window

Modalities

Reasoning modes

Integration

Reasoning trade-offs

Source (opens in a new tab)
Intelligence Index v4.3.2
DeepSeek-V4-Flash-0731
Cost per task (USD) · log scale
DeepSeek-V4-Flash-0731Max
Intelligence Index
34.3
Cost / task
$0.2196
Evaluation time
3.4m

Pricing & access

API route
OpenRouter · CloudflareAPI
Tier & contextStandard

USD per 1M tokens

Input
$0.44
Output
$1.32
Cached input
$0.014

Run locally

GGUFMXFP4
Context
Estimated memory

~160.6 GiB

RAM + VRAM

Estimate assumptions

One text chat at the selected context, with a 16-bit cache and app allowance. Operating system and image processing are extra; optimized caches can use less.

192 GiB RAMwithout GPU offload

Memory budgetOne text chat
Weights
145.6 GiB
Context + state
0.4 GiB16-bit cache
App allowance
14.6 GiB
Download
156.4 GB4 files · keep together
Disk space
~178 GB
Available software
Ollama (opens in a new tab)LM Studio (opens in a new tab)Jan (opens in a new tab)KoboldCpp (opens in a new tab)llama.cpp (opens in a new tab)
Get GGUF (opens in a new tab)
Set up in Ollama · MXFP4 · 8K context

Save the following as Modelfile. The context matches the estimate above.

FROM hf.co/lmstudio-community/DeepSeek-V4-Flash-0731-GGUF:MXFP4
PARAMETER num_ctx 8192

Then run these commands in your terminal:

ollama create llmcurrent-local -f Modelfile
ollama run llmcurrent-local

Ollama import instructions (opens in a new tab) · For LM Studio or Jan, open the app’s model search, enter lmstudio-community/DeepSeek-V4-Flash-0731-GGUF, choose MXFP4, and set context to 8,192 tokens in the app’s model settings. Software links above open each app’s setup guide.

vLLM setup (opens in a new tab)

Benchmarks

Full leaderboardSource (opens in a new tab)

Humanity’s Last Exam

Frontier reasoning

DeepSeek-V4-Flash-0731 · MaxArchived

  1. 1Claude Opus 5.5Max61.4%
  2. 2Claude Fable 5.1Max59.1%
  3. 3Gemini 4 ArgonHigh57.1%
  4. 4Claude Sonnet 5.5Max55.0%
  5. 5GPT-6 AstraMax54.7%
  6. 6Muse Wonder FullMax53.6%
  7. 7GPT-6.1 SolMax52.9%
  8. 8Muse WonderMax50.1%
  9. 9MiMo-V2.6-ProReasoning49.4%
  10. 10Muse Spark 1.3Max48.7%

Peers use their best measured result on this benchmark.