whatmodel.sh / DeepSeek V4.1 Flash

DeepSeek V4.1 Flash

+ contribute

DeepSeek V4.1 Flash is an open-weights language model from DeepSeek, released September 2026. Its published size is 552B parameters (16B active per token) and a 1M-token context window. Licence: mit.

77.5±0.7#21 of 719 LLMsfrom 24 benchmark results across 4 sources

$ how it compares

The LLMs ranked either side of DeepSeek V4.1 Flash, out of 828 with a score, then more from DeepSeek.

#	model	score	where	RAM	released
19	Claude Opus 4.8 Anthropic	78±0	api $10.00/M		2026-05
20	GPT-5.4 OpenAI	78±1	api $5.63/M		2026-03
21	DeepSeek V4.1 Flash DeepSeek	78±1	api $0.26/M +local	~537GB	2026-09
22	Claude Haiku 5.5 Anthropic	77±1	api $0.20/M		2026-10
23	Gemini 3.8 Flash Google	77±1	api $1.50/M		2026-09
	more from DeepSeek
31	DeepSeek V4 Pro (0813) DeepSeek	76±1	api $1.63/M +local	~1010GB	2026-08
35	DeepSeek V3.2 Speciale DeepSeek	76±3	comparison only
36	DeepSeek V4 Flash Vision (Exp) DeepSeek	76±2	local	~104GB	2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #12 of 300 · 21% of its score · 9 tests

Beats GPT-5.5 Pro, GPT-5.4 Pro, GPT-5.5 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
Arena Agent (October 2026 aggregate IPS)	0.0402 IPS ✎	▁▁▁▄▂▃▇█▂▂▁▃ beats 34/49	avg -0.312 IPS	▇▇▇ 8.3%	sharp
Terminal-Bench 4.0 lab	31.2% ✎	▁·▁··▁▇▇▅██▃ beats 16/30	very hard · avg 0.3%	▇ 3.5%	sharp · small run
Terminal-Bench 2.1 lab	90.6% ✎	▁·▁·▁▁▄▃▄█▃▇ beats 55/56	hard · avg 26%	▇ 2.8%	small run
Video2World V2WScore (coding agents reconstruct interactive simulated worlds from videos, OpenCode scaffold)	30.5% ✎	█··▄█·▄····▄ beats 3/6	very hard · avg 6.6%	▇ 2.0%	at its level
Humanity's Last Exam with tools lab	63.9% ✎	▂▂▂▁▄▂▃▆█▇▂▆ beats 49/54	hard · avg 17%	▇ 1.4%	at its level · small run
DeepSWE v1.1 (resolved) lab	74.2% ✎	▂····▁▁▁▆██▆ beats 27/28	hard · avg 21%	▇ 1.1%	at its level · small run
Agents' Last Exam (pass@1, official scaffold) lab	31.8% ✎	█··········▄ beats 2/2	very hard · avg 6.2%	▇ 1.1%	at its level · small run
Terminal-Bench 3.0 lab	30.0% ✎	█··█·······█ beats 2/2	very hard · avg 2.1%	▇ 0.7%	noisy · small run

1 more test

test	result	vs the others in its run	difficulty	counts	why
Chartography with tools lab	78.9% ✎	▃········▅██ beats 2/8		—	not counted yet: no other model in the same run to compare it with

coding▁▁▁▃▆██▇▆▄▃▁ #22 of 458 * · 8.5% of its score · 2 tests

Beats GPT-5.5, MiMo-V2.6-Pro, GPT-5.6 Terra here, though they rank higher overall. Trails Grok 4.7, Qwen3.8-Flash-Next, Qwen3.8 Max (0902) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Arena WebDev (Code Arena)	1619.9 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 101/119	avg 981	▇▇▇ 7.3%	sharp
DeepSWE v1.1 (resolved) lab	74.2% ✎	▂····▁▁▁▆██▆ beats 27/28	hard · avg 21%	▇ 1.1%	at its level · small run

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #27 of 393 * · 4.1% of its score · 1 test

Beats Claude Sonnet 5.5, GPT-6 Sol, GPT-5.6 Terra here, though they rank higher overall. Trails GLM-5.2, Gemini 3.6 Flash, Gemini 3.5 Flash here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1474.4 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 349/379	avg 1268	▇▇ 4.1%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #32 of 398 · 38% of its score · 4 tests

Trails GPT-5.2, Qwen3.7 Max, Gemini 3.1 Pro (Preview) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
FrontierMath tier 4 v2	26.8% ✎	▁▂▁▂█▅▂▂▁▁▁▁ beats 28/63	very hard · avg 0.1%	▇▇▇▇▇▇▇▇ 19%	sharp · at its level
FrontierMath tiers 1–3 v2	67.4% ✎	▂··▁▁▃▄▄█▅▃▃ beats 55/82	very hard · avg 1.2%	▇▇▇▇▇▇ 16%	sharp · at its level
MathArena Apex (pass@1) lab	65.6% ✎	▄····▄···▄█▄ beats 4/5	very hard · avg 15%	▇ 1.8%	at its level · small run
OTIS mock AIME 2024–2025	98.3% ✎	▃▃▄▃▃▃▇▇█▃▃▅ beats 173/193	very hard · avg 8.5%	▇ 0.5%	sharp · maxed out · noisy

web search██·········█ #1 of 3 * · 1.4% of its score · 1 test

test	result	vs the others in its run	difficulty	counts	why
Humanity's Last Exam with tools lab	63.9% ✎	▂▂▂▁▄▂▃▆█▇▂▆ beats 49/54	hard · avg 17%	▇ 1.4%	at its level · small run

other tests12 not in a task · 30% of its score

test	result	vs the others in its run	difficulty	counts	why
Furniture Assembly (Epoch)	34.2% ✎	█▆▆▆▆▂▅▂▂▂·▅ beats 11/30	very hard · avg 1.2%	▇▇▇ 7.5%	sharp · at its level
Mystery Game Puzzles (Epoch)	43.0% ✎	█··▂▃▇▄█▃▃▁▁ beats 62/74	very hard · avg 9.4%	▇▇▇ 6.7%	sharp · at its level · noisy
GPQA Diamond +1	89.8% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 172/208	hard · avg 39%	▇▇▇ 6.4%	sharp · small run · several runs
Humanity's Last Exam lab	39.1% ✎	▁▁▁▃▆▆▇██▆▁▃ beats 88/104	very hard · avg 5.3%	▇ 2.4%	at its level · small run
NL2Repo lab	64.0% ✎	▂▂·▂▂▄█▆▆▂▄▆ beats 21/21	very hard · avg 13%	▇ 2.0%	at its level · small run
SEC-Bench Pro (pass@1) lab	62.8% ✎	▄··▄···▄·▄·█ beats 4/5	very hard · avg 7.2%	▇ 1.5%	at its level · small run
CyberGym lab	88.1% ✎	▄···▄··▄█▄█▄ beats 8/8	medium · avg 36%	▇ 1.2%	small run
ExploitGym (pass@1) lab	15.3% ✎	▄···▄··▄▄··█ beats 4/5	very hard · avg 0.5%	▇ 1.1%	noisy · small run

4 more tests

test	result	vs the others in its run	difficulty	counts	why
AutomationBench (DeepSeek V4.1 card) lab	54.8% ✎	█··█·······█ beats 2/2	very hard · avg 12%	▇ 1.0%	at its level · small run
ProgramBench (Almost@1) lab	20.3% ✎	█··········█ beats 1/1		—	not counted yet: no other model in the same run to compare it with
BabyVision with tools lab	89.6% ✎	no other model yet		—	not counted yet: no other model in the same run to compare it with
ZeroBench main with tools (pass@5) lab	49.0% ✎	no other model yet		—	not counted yet: no other model in the same run to compare it with

$ where to run it

Through an API

api · $0.26/M · $0.15 in / $0.6 out per 1M tokens · 1M context · api-docs.deepseek.com ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ.

Other ways to run it

Run it as published · needs ~510.3 GB, NVIDIA only · from huggingface.co/deepseek-ai
hf download deepseek-ai/DeepSeek-V4.1-Flash --local-dir DeepSeek-V4.1-Flash-HF && cd DeepSeek-V4.1-Flash-HF/inference && python -m pip install -r requirements.txt && python convert.py --hf-ckpt-path .. --save-path ../../DeepSeek-V4.1-Flash-TP8 --model-parallel 8 --expert-dtype fp4 --tokenizer-path .. && torchrun --nproc-per-node 8 generate.py --ckpt-path ../../DeepSeek-V4.1-Flash-TP8 --config config.json --interactive --temperature 0.6

Weights and documentation: huggingface.co/deepseek-ai ↗

$ how the score was worked out · Epoch AI — internal evaluatio… 52%, deepseek-ai/DeepSeek-V4.1-Fla… 26%, LMArena leaderboard dataset 20% and 1 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
Epoch AI — internal evaluations	6	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 52%	CC BY 4.0
deepseek-ai/DeepSeek-V4.1-Flash model card · lab-reported	14	▇▇▇▇▇▇▇▇ 26%	MIT
LMArena leaderboard dataset	3	▇▇▇▇▇▇ 20%	CC BY 4.0
Video2World: Benchmarking Coding Agents for Interactive World Modeling from Embodied Videos (arXiv:2610.04432v2)	1	▇ 2%	CC BY-SA 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 77.5

not counted (4)

deepseek-ai/DeepSeek-V4.1-Flash model card	BabyVision with tools [correct]	89.6%	fewer than three models have this test yet, or this source ran it for this model alone
deepseek-ai/DeepSeek-V4.1-Flash model card	Chartography with tools [correct]	78.9%	fewer than three models have this test yet, or this source ran it for this model alone
deepseek-ai/DeepSeek-V4.1-Flash model card	ProgramBench (Almost@1) [correct]	20.3%	fewer than three models have this test yet, or this source ran it for this model alone
deepseek-ai/DeepSeek-V4.1-Flash model card	ZeroBench main with tools (pass@5) [correct]	49.0%	fewer than three models have this test yet, or this source ran it for this model alone

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated 2026-10-11 02:20 UTC · sources and licences · json