Skip to content
AIMODELRACE
DeepSeekLanguage modelOpen weightsNew

DeepSeek V4.1 Flash

MIT-licensed speed demon: near-frontier coding for pennies, with built-in search.

Compare head-to-headRace it in 3D

Overall Intelligence

#35

of 56
14.1 pts behind #1

0

races led

0

weeks at #1

74.6

score

Released
Oct 17, 2026
2 days ago
Context
1M
128K max output
Price
$0.3 / $1.2
per 1M in/out · $0.52 blended
Speed
182 tok/s
1.2s to first token
Modalities
text
outputs text
Weights
Open
552B MoE (8B active)

The 30-second version

An MIT-licensed speed demon: near-frontier coding for pennies, with built-in search.

Key takeaways
  • 1$0.30 / $1.20 per million tokens.
  • 2Fast, at around 180 tokens per second.
  • 3MIT licence: about as permissive as it gets.
  • 4A great budget coding assistant.

What you need to know

Purpose

Cheap, fast and open intelligence for coding and search-augmented answers.

What DeepSeek optimised for

Speed and efficiency on top of the V4 base, with search built in.

Unique strengths

  • MIT licence.
  • Built-in web search.
  • Excellent value for coding.
Insight
How it differs from rivals. Versus Gemini 3.5 Flash: open and cheaper on output, slower and weaker multimodally.

Where it ranks

Current snapshot · illustrative scores · click any race to watch it

CodingHard ReasoningMathScienceWriting & LanguageMultimodalTool Use & AgentsLong ContextTrust & FactualitySpeed & Cost

V4.1 FlashBest on each axis

Overall Intelligence over time

#1 frontierV4.1 FlashGLM-5.3Large 4
55.060.065.070.075.080.085.090.02025Apr ’25Jul ’25Oct ’252026Apr ’26Jul ’26Oct ’26

Best use cases

01

Budget coding assistant

Autocomplete and quick fixes at scale.

02

Search-backed Q&A

Current answers with citations, cheaply.

Strengths

  • Value

    One of the best quality-per-dollar models in the race.

Weaknesses

  • Depth

    Not for hard reasoning or long agent runs.

When to choose it, and when not to

V4.1 Flashover3.5 Flash

You want open weights and a permissive licence.

Pick something else when…

  • Multimodal inputs.
  • High-stakes reasoning.

Update history

The distilled version first; open any release for the deeper breakdown.