DeepSeek V4.1 Flash

Sep 2026
Older versionReplace withDeepSeek V4.1 Flash 1017Newer versionGPT-6.1 SolCheaper, as goodClaude Haiku 5.5BetterMiMo v2.6 ProOpen weights
DeepSeek V4.1 Flash
Compare in
on
$0.49per task?
$0.30 / $1.20per 1M in / out?
174 tok/sfirst token in 0.8 s?
1Mcontext

Rankingat max effort?

Retired models are left out of the current rankings.

Alternatives?

ModelScoreCostPer taskWhat changes
CheaperGPT-6.1 Solxhigh+29%-19%$0.39Remove unsupported parameters+5OpenMiMo v2.6 Prothinking+17%-73%$0.13Pass back reasoning_content+3CheaperDeepSeek V4.1 Flash 1017maxOpen+14%-46%$0.26Pass back reasoning_content · Chat Completions+4CheaperClaude Haiku 5.5max+10%-56%$0.21Remove sampling parameters+5CheaperMiniMax M3.1 Flashthinking+8%-55%$0.22Complete model response must be appended

Run it yourself

Fits onhardware needs not verified yet
Licencenot verified

History

4 versions
2530354045Jul 26Oct 26Release dateAA Intelligence Index 4.3.2V4 · AprV4 · JulV4.1 · SepV4.1 · Oct
2530354045Jul 26Oct 26Release dateAA Intelligence Index 4.3.2V4 · AprV4 · JulV4.1 · Sep

Benchmarks

all categories
Chat & Knowledge
Humanity's Last Exam39.2
Arena Text · Chinese1,503.3
Arena Text · English1,473.1
Arena Text · French1,463.1
Arena Text · German1,476
Arena Text · instruction following1,473.4
Arena Text · Japanese1,420.6
Arena Text · Korean1,440.6
Arena Text · non-English1,448.7
Arena Text · Polish1,453.4
Arena Text · Russian1,465.4
Arena Text · Spanish1,458.6
AA Omniscience Index-5.3
SimpleQA Verified v142.9
CritPt (physics)14.3
AA Intelligence Index 4.3.239.5
AA Omniscience · Accuracy46.4
AA Omniscience · Hallucination Rate96.5
AA-LCR (long context) 1.184
GDP.pdf · AA all-pass12.8
MLCR-AA22.8
ARC-AGI-1 · public evaluation98
ARC-AGI-1 · semi-private94.5
ARC-AGI-2 · public evaluation81.7
ARC-AGI-2 · semi-private72.9
ARC-AGI-3 · semi-private0.9
FrontierMath Tiers 1–3 (v2)67.4
FrontierMath Tier 4 (v2)26.8
Chess Puzzles48
Mystery Game Puzzles43
MathArena · ArXivMath · April 202645
MathArena · ArXivMath · August 202643.9
MathArena · ArXivMath · June 202657.6
MathArena · ArXivMath · May 202658.3
MathArena · ArXivMath · September 202640.6
MathArena · BrokenArXiv · August 202654.8
MathArena · BrokenArXiv · June 202639.4
MathArena · BrokenArXiv · May 202628.5
MathArena · BrokenArXiv · September 202654.7
MathArena · ArXivLean · September 20265.7
Arena Text · overall · full cohort1,462.8
GPQA Diamond · Epoch89.8
OTIS Mock AIME 2024 202598.3
Coding
Terminal-Bench 4.026.8
FrontierCode 1.144
SciCode51.9
Terminal-Bench Science 0.19
Agents & Tool Use
AutomationBench-AA68.9
GDPval-AA (work tasks) 2.155
AA Analyst Agent35
ITBench SRE46.9
AA-Briefcase 1.1 · Elo1,419.4
Harvey LAB-AA 1.11.7
Furniture Assembly34.2
Writing & Language
Arena Text · creative writing · full cohort1,436.5
Vision & Documents
MMMU-Pro (vision)77
Chartography11.6

FAQ

How much does DeepSeek V4.1 Flash cost?›

$0.30 per 1M input tokens and $1.20 per 1M output tokens. A typical benchmark task costs about $0.49 at max effort.

Is DeepSeek V4.1 Flash free to use?›

There is no free API; the cheapest paid price is $0.30 per 1M input tokens. The weights are open, so you can run it yourself.

How can I use DeepSeek V4.1 Flash?›

Through the DeepSeek API.

What is the context window of DeepSeek V4.1 Flash?›

1M tokens.

When was DeepSeek V4.1 Flash released? Is there a newer version?›

Released 2026-09-10. It has been superseded by DeepSeek V4.1 Flash 1017.

Is DeepSeek V4.1 Flash open source?›

Yes, the weights are open.

Can I run DeepSeek V4.1 Flash locally?›

The weights are open and can be self-hosted; hardware needs are not confirmed yet.

What is DeepSeek V4.1 Flash?›

DeepSeek V4.1 Flash is an open-weights language model from DeepSeek, released Sep 2026, the successor to DeepSeek V4 Flash 0731.

What are the best alternatives to DeepSeek V4.1 Flash?›

GPT-6.1 Sol (cheaper), MiMo v2.6 Pro (open weights).