Ctrl K
Back to models

DeepSeek

DeepSeek-V4.1-Flash

  • Released
  • Open weights
  • MIT
Compare
Intelligence Index

Max setting

39.5#23 of 83measured models
Measured · Artificial Analysis v4.3.2
Context window
1.05M tok

Performance profile

CodingRank #27 of 94
51.9%
SciCode
Cost efficiencyRank #19 of 67
148.8pts / $
Intelligence ÷ eval cost

Capabilities

Context window

Modalities

Reasoning modes

Avg reasoning tokens / task — Artificial Analysis Intelligence Index v4.3.2: weighted average reasoning tokens per benchmark task, not a token limit. Source (opens in a new tab)

Reasoning trade-offs

Source (opens in a new tab)
Intelligence Index v4.3.2
DeepSeek-V4.1-Flash
Cost per task (USD) · log scale
DeepSeek-V4.1-FlashMax
Intelligence Index
39.5
Cost / task
$0.2652
Evaluation time
6.0m

Pricing & access

API route
DeepSeekRetired
Tier & context

USD per 1M tokens

Input
$0.30
Output
$1.20
Cached input
$0.006

Run locally

GGUFMXFP4
8K context · text only
Tested desktop setup
System RAM125.7 GiB
GPU memory31.8 GiB

RTX 5090 · reference setup, not a minimum

Memory budget157.5 GiB installed
RAM cache
72 GiBModel weights
GPU cache
18 GiBModel weights
Other memory
67.5 GiBSystem + model + spare
Download
501.8 GB11 files · keep together
Disk space
~557 GBPCIe 5 NVMe · disk streaming
Available software
llama.cpp (opens in a new tab)
Custom build · setup guide (opens in a new tab)
Get GGUF (opens in a new tab)

Benchmarks

Full leaderboardSource (opens in a new tab)

Humanity’s Last Exam

Frontier reasoning

DeepSeek-V4.1-Flash · Max#27 of 126

  1. 1Claude Opus 5.5Max61.4%
  2. 2Claude Fable 5.1Max59.1%
  3. 3Gemini 4 ArgonHigh57.1%
  5. 24DeepSeek-V4-Pro-0813Max41.0%
  6. 25GLM-5.3-FlashMax39.9%
  7. 26Ling-3.1-flashReasoning39.4%
  8. 27DeepSeek-V4.1-FlashMax39.2%
  9. 28MiniMax-M3Reasoning39.0%
  10. 29GPT-6 LunaMax38.5%
  11. 30Agnes 3.0 FlashReasoning38.5%

Peers use their best measured result on this benchmark.

Alternatives

Each model at its highest-scoring measured setting · Intelligence Index v4.3.2 and its evaluation cost per task.

Inspect conflicting source claims

Some sources disagree. The selected claim follows the catalog’s source precedence.

DeepSeek-V4.1-Flash — direct API text inference

Selected claim · reviewed-endpoint-correctionsOriginal source (opens in a new tab)
{
  "api_model_id": "deepseek-flash",
  "context_window": 1000000,
  "lifecycle": "retired",
  "max_output_tokens": 384000,
  "name": "DeepSeek-V4.1-Flash — direct API text inference",
  "region": null,
  "supports_json_schema": null,
  "supports_tools": true,
  "url": "https://api.deepseek.com"
}
Alternate claim · deepseek-pricingOriginal source (opens in a new tab)
{
  "api_model_id": "deepseek-flash",
  "context_window": 1000000,
  "lifecycle": "active",
  "max_output_tokens": 384000,
  "name": "DeepSeek-V4.1-Flash — direct API text inference",
  "region": null,
  "supports_json_schema": null,
  "supports_tools": true,
  "url": "https://api.deepseek.com"
}