DeepSeek V4.1 Flash benchmarks, price, and specs

AA Intelligence: 39#24 of 142
Input / output per 1M: $0.30/ $1.20
Output speed: 235tok/s
Context window: 1M tokens

Composite indexes

Artificial Analysis Intelligence Index#24 of 142
39independentv4.3.2, reasoning, max effortArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
25independentv4.3.2, non-reasoningArtificial Analysis: DeepSeek V4.1 Flash (Non-Reasoning) ↗
Leader: Claude Opus 5.5 at 58

Coding

Terminal-Bench 4.0#17 of 88
31.2%vendormax effort, DeepSeek Harness (Minimal mode), 1M contextDeepSeek-V4.1-Flash model card (Hugging Face) ↗
26.8%independentreasoning, max effort, AA-runArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: Claude Mythos 5.1 at 60.9%
SciCode#32 of 94
51.9%independentreasoning, max effort, AA-runArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: Claude Opus 5.5 at 66.9%
Terminal-Bench 2.x#2 of 103
90.6%vendorTerminal-Bench 2.1, max effort, DeepSeek Harness (Minimal mode)DeepSeek-V4.1-Flash model card (Hugging Face) ↗
Leader: Claude Fable 5.1 at 91.4%

Agents and tool use

GDPval-AA#12 of 103
1600independentGDPval-AA v2.1, reasoning, max effortArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: Claude Opus 5.5 at 1846

Reasoning and knowledge

Humanity’s Last Exam#37 of 147
63.9%vendormax effort, with toolsDeepSeek-V4.1-Flash model card (Hugging Face) ↗
39.2%independentreasoning, max effort, no tools, AA-runArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
36.8%vendormax effort, no tools, full set (39.1 on text-only subset)DeepSeek-V4.1-Flash model card (Hugging Face) ↗
Leader: Claude Opus 5.5 at 61.4%
CritPt#33 of 139
14.3%independentreasoning, max effort, AA-runArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: GPT-5.6 Sol at 32.3%
GPQA Diamond#34 of 142
90.9%vendormax reasoning effort (reasoning_effort=100), Pass@1DeepSeek-V4.1-Flash model card (Hugging Face) ↗
Leader: GPT-6 Astra at 96.1%

Vision and long context

AA Long Context Reasoning#7 of 139
84%independentreasoning, max effort, AA-LCRArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: Kimi K3 at 88.7%
MMMU-Pro#29 of 72
77%independentreasoning, max effort, AA-runArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: Claude Opus 5.5 at 87.7%

Human preference

LMArena WebDev#15 of 54
1620independentarena entry "deepseek-v4.1-flash-max", 3,265 votes, checked 2026-09-25Arena (LMArena) WebDev/Code leaderboard ↗
Leader: Claude Opus 5.5 at 1818

Other published results

Result	Score	Reported by	Source
Terminal-Bench 3.0 · max effort, DeepSeek Harness (Minimal mode)	30%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
DeepSWE v1.1 · max effort, mini-SWE harness	74.2%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
Agents' Last Exam · max effort, official scaffold	31.8%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
AutomationBench · max effort, official scaffold	54.8%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
NL2Repo-Bench · max effort	64 score	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
CyberGym · max effort	88.1%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
Codeforces rating · max effort	3471 Elo	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
MathArena Apex · max effort	65.6%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗

Specs

Released: September 10, 2026
Parameters: 552B total, 16B active
Size class: 500B to 1T
Input: text, image
Output: text
Context: 1M tokens
Max output: 384K tokens
Cached input: $0.006 / 1M
Blended 70/30: $0.57 / 1M
First token: 0.97s

Price: DeepSeek API: Models & Pricing ↗ API id deepseek-flash. Peak-hour list rates (peak = 01:00-04:00 and 06:00-10:00 UTC, Mon-Fri). Off-peak rates are half: $0.15 in / $0.60 out / $0.003 cache hit. Legacy ids deepseek-v4-flash and deepseek-v4-flash-vision-exp are now served by this model at these prices.
Speed: Artificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) (max effort) ↗

Will it fit your budget?

The planner prices this model against local hardware and other APIs for your workload.

Plan a stack →

Compare with

AA Intelligence: 39#24 of 142
Input / output per 1M: $0.30/ $1.20
Output speed: 235tok/s
Context window: 1M tokens

Composite indexes

Artificial Analysis Intelligence Index#24 of 142
39independentv4.3.2, reasoning, max effortArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
25independentv4.3.2, non-reasoningArtificial Analysis: DeepSeek V4.1 Flash (Non-Reasoning) ↗
Leader: Claude Opus 5.5 at 58

Coding

Terminal-Bench 4.0#17 of 88
31.2%vendormax effort, DeepSeek Harness (Minimal mode), 1M contextDeepSeek-V4.1-Flash model card (Hugging Face) ↗
26.8%independentreasoning, max effort, AA-runArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: Claude Mythos 5.1 at 60.9%
SciCode#32 of 94
51.9%independentreasoning, max effort, AA-runArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: Claude Opus 5.5 at 66.9%
Terminal-Bench 2.x#2 of 103
90.6%vendorTerminal-Bench 2.1, max effort, DeepSeek Harness (Minimal mode)DeepSeek-V4.1-Flash model card (Hugging Face) ↗
Leader: Claude Fable 5.1 at 91.4%

Agents and tool use

GDPval-AA#12 of 103
1600independentGDPval-AA v2.1, reasoning, max effortArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: Claude Opus 5.5 at 1846

Reasoning and knowledge

Humanity’s Last Exam#37 of 147
63.9%vendormax effort, with toolsDeepSeek-V4.1-Flash model card (Hugging Face) ↗
39.2%independentreasoning, max effort, no tools, AA-runArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
36.8%vendormax effort, no tools, full set (39.1 on text-only subset)DeepSeek-V4.1-Flash model card (Hugging Face) ↗
Leader: Claude Opus 5.5 at 61.4%
CritPt#33 of 139
14.3%independentreasoning, max effort, AA-runArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: GPT-5.6 Sol at 32.3%
GPQA Diamond#34 of 142
90.9%vendormax reasoning effort (reasoning_effort=100), Pass@1DeepSeek-V4.1-Flash model card (Hugging Face) ↗
Leader: GPT-6 Astra at 96.1%

Vision and long context

AA Long Context Reasoning#7 of 139
84%independentreasoning, max effort, AA-LCRArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: Kimi K3 at 88.7%
MMMU-Pro#29 of 72
77%independentreasoning, max effort, AA-runArtificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) ↗
Leader: Claude Opus 5.5 at 87.7%

Human preference

LMArena WebDev#15 of 54
1620independentarena entry "deepseek-v4.1-flash-max", 3,265 votes, checked 2026-09-25Arena (LMArena) WebDev/Code leaderboard ↗
Leader: Claude Opus 5.5 at 1818

Other published results

Result	Score	Reported by	Source
Terminal-Bench 3.0 · max effort, DeepSeek Harness (Minimal mode)	30%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
DeepSWE v1.1 · max effort, mini-SWE harness	74.2%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
Agents' Last Exam · max effort, official scaffold	31.8%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
AutomationBench · max effort, official scaffold	54.8%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
NL2Repo-Bench · max effort	64 score	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
CyberGym · max effort	88.1%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
Codeforces rating · max effort	3471 Elo	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗
MathArena Apex · max effort	65.6%	vendor	DeepSeek-V4.1-Flash model card (Hugging Face) ↗

Specs

Released: September 10, 2026
Parameters: 552B total, 16B active
Size class: 500B to 1T
Input: text, image
Output: text
Context: 1M tokens
Max output: 384K tokens
Cached input: $0.006 / 1M
Blended 70/30: $0.57 / 1M
First token: 0.97s

Price: DeepSeek API: Models & Pricing ↗ API id deepseek-flash. Peak-hour list rates (peak = 01:00-04:00 and 06:00-10:00 UTC, Mon-Fri). Off-peak rates are half: $0.15 in / $0.60 out / $0.003 cache hit. Legacy ids deepseek-v4-flash and deepseek-v4-flash-vision-exp are now served by this model at these prices.
Speed: Artificial Analysis: DeepSeek V4.1 Flash (Reasoning, Max Effort) (max effort) ↗

Will it fit your budget?

The planner prices this model against local hardware and other APIs for your workload.

Plan a stack →