DeepSeek V4.1 Flash 1017 — Benchmarks

Benchmark scores for DeepSeek V4.1 Flash 1017 aggregated from public leaderboards (15 boards), with how it ranks among open models and how it compares with its nearest neighbours. See hardware requirements for what you need to run it.

llmrun Composite Score →

133.2

90% range 130.6–135.8 · from 10 boards

CodingNot enough data yet
AgentsNot enough data yet
Math132.090% range 129–135
ScienceNot enough data yet
Reasoning139.890% range 133–146

DeepSeek V4.1 Flash 1017 reasoning benchmarks

Multi-step logic and problem-solving — can the model think through a hard problem rather than recall an answer.

GPQA Diamond
GPQA Diamond is a set of extremely hard, graduate-level science questions (physics, chemistry, biology) written by domain experts and filtered so that skilled non-experts with web access still fail. It measures genuine reasoning rather than memorization.
93.2%
93.2%
median 55.6%open leader
Overall best: Gemini 4 Argon (high) 96.4%
top · #1 of 99 open · #20 of 301 all
#1 of 99
#20 of 301
Independent
Epoch AI · Oct 2026
Chess Puzzles
Chess Puzzles is an Epoch AI evaluation that asks models to find the winning move in tactical chess positions drawn from real games. Because each puzzle has one verifiable correct answer and no natural-language shortcut, it isolates concrete multi-step planning.
46.0%
46.0%
median 21.0%open leader 48.0%
Overall best: GPT 6 Astra Promax 76.0%
89th percentile · #2 of 15 open · #26 of 147 all
#2 of 15
#26 of 147
Independent
Epoch AI · Oct 2026
LiveBench 2026 Reasoning
LiveBench 2026 Reasoning measures logical, multi-step problem solving. It belongs to the 2026 LiveBench release (first snapshot September 2, 2026), a redesigned LiveBench in which each of seven categories is scored by a single task and questions are refreshed to limit training-set contamination. It is a different test from the original LiveBench boards, so the scores are not comparable.
89.1
89.1
median 78.8open leader 90.7
Overall best: Gemini 4 Argon 95.1
94th percentile · #2 of 17 open · #20 of 53 all
#2 of 17
#20 of 53
Independent
LiveBench · Oct 2026
Mystery Game Puzzles
Mystery Game Puzzles is an Epoch AI evaluation of multi-step deductive reasoning: the model works through a mystery or logic-game scenario to arrive at the correct solution. It probes sustained, structured inference rather than single-step question answering.
49.0%
49.0%
median 18.0%open leader
Overall best: GPT 6 Astra Max 84.0%
top · #1 of 19 open · #14 of 134 all
#1 of 19
#14 of 134
Independent
Epoch AI · Oct 2026

Other results (2)· leaderboard-only, not part of the llmrun Score

LiveBench 2026 Browser
LiveBench 2026 Browser measures tasks a model completes by operating a web browser. It belongs to the 2026 LiveBench release (first snapshot September 2, 2026), a redesigned LiveBench in which each of seven categories is scored by a single task and questions are refreshed to limit training-set contamination. It is a different test from the original LiveBench boards, so the scores are not comparable.
76.0
76.0
median 16.0open leader 92.0
Overall best: Claude Fable 5.1 100.0
75th percentile · #5 of 17 open · #20 of 53 all
#5 of 17
#20 of 53
Independent
LiveBench · Oct 2026
LiveBench 2026 Data Analysis
LiveBench 2026 Data Analysis measures questions answered by working with tables and structured data. It belongs to the 2026 LiveBench release (first snapshot September 2, 2026), a redesigned LiveBench in which each of seven categories is scored by a single task and questions are refreshed to limit training-set contamination. It is a different test from the original LiveBench boards, so the scores are not comparable.
79.5
79.5
median 72.3open leader
Overall best: Gemini 4 Argon 85.2
top · #1 of 17 open · #12 of 53 all
#1 of 17
#12 of 53
Independent
LiveBench · Oct 2026

DeepSeek V4.1 Flash 1017 coding benchmarks

Writing, editing, and fixing real code — measured by whether the resulting program actually runs and passes tests.

LiveBench 2026 Coding
LiveBench 2026 Coding measures whether a model writes correct code for programming problems. It belongs to the 2026 LiveBench release (first snapshot September 2, 2026), a redesigned LiveBench in which each of seven categories is scored by a single task and questions are refreshed to limit training-set contamination. It is a different test from the original LiveBench boards, so the scores are not comparable.
96.0
96.0
median 73.6open leader
Overall best: Claude Fable 5 98.3
top · #1 of 17 open · #4 of 53 all
#1 of 17
#4 of 53
Independent
LiveBench · Oct 2026
LiveBench 2026 Long-Horizon
LiveBench 2026 Long-Horizon measures multi-step tasks that a model has to carry through many steps to finish. It belongs to the 2026 LiveBench release (first snapshot September 2, 2026), a redesigned LiveBench in which each of seven categories is scored by a single task and questions are refreshed to limit training-set contamination. It is a different test from the original LiveBench boards, so the scores are not comparable.
50.9
50.9
median 28.4open leader
Overall best: Claude Opus 5.5 85.0
top · #1 of 17 open · #12 of 53 all
#1 of 17
#12 of 53
Independent
LiveBench · Oct 2026

Other results (2)· leaderboard-only, not part of the llmrun Score

LMArena WebDev
Code Arena (formerly WebDev Arena) is Arena's head-to-head coding leaderboard: two models each build a real web app or UI from the same prompt, and people vote blind on which result is better. The rating shown is a relative Bradley-Terry score, not a percentage — compare it only to other models on this board.
1665.0
1665.0
median 1414.9open leader
Overall best: Claude Opus 5.5 (Max) 1806.9
top · #1 of 55 open · #14 of 144 all
#1 of 55
#14 of 144
Arena
Arena · Oct 2026
LiveBench 2026 Design
LiveBench 2026 Design measures layout-design tasks (LiveBench's design_layout task). It belongs to the 2026 LiveBench release (first snapshot September 2, 2026), a redesigned LiveBench in which each of seven categories is scored by a single task and questions are refreshed to limit training-set contamination. It is a different test from the original LiveBench boards, so the scores are not comparable.
83.4
83.4
median 71.7open leader
Overall best: Claude Opus 5.5 86.9
top · #1 of 17 open · #5 of 53 all
#1 of 17
#5 of 53
Independent
LiveBench · Oct 2026

DeepSeek V4.1 Flash 1017 math benchmarks

Symbolic and quantitative problem-solving, from competition math to multi-step calculation.

AIME 2024/2025
AIME (American Invitational Mathematics Examination) is a prestigious high-school competition of hard, integer-answer problems. It's a widely-cited yardstick for multi-step mathematical reasoning, here on the 2024–2025 papers.
99.4%
99.4%
median 53.3%open leader
Overall best: Claude Fable 5 (high) 100.0%
top · #1 of 87 open · #15 of 284 all
#1 of 87
#15 of 284
Independent
Epoch AI · Oct 2026
FrontierMath
FrontierMath is a benchmark of exceptionally hard, original research-level mathematics problems created with professional mathematicians. Even the strongest models solve only a small fraction, making it a frontier measure of genuine mathematical ability.
71.2%
71.2%
median 56.5%open leader 72.2%
Overall best: GPT 6 Astra Max 93.7%
94th percentile · #2 of 18 open · #27 of 115 all
#2 of 18
#27 of 115
Independent
Epoch AI · Oct 2026
FrontierMath Tier 4
FrontierMath Tier 4 is the hardest tier of Epoch AI's research-level mathematics benchmark, holding problems that working mathematicians consider genuinely difficult research questions. Scores here stay low even for frontier models, making it the sharpest available read on mathematical ability.
46.3%
46.3%
median 26.2%open leader
Overall best: GPT 6.1 Sol Max 100.0%
top · #1 of 14 open · #28 of 77 all
#1 of 14
#28 of 77
Independent
Epoch AI · Oct 2026

DeepSeek V4.1 Flash 1017 knowledge benchmarks

Breadth and depth of factual knowledge across many subjects, paired with the reasoning to apply it.

SimpleQA
SimpleQA measures factual accuracy on short, fact-seeking questions with a single correct answer — directly probing how often a model is right versus confidently wrong (hallucination) on simple facts.
42.5%
42.5%
median 38.4%open leader 56.6%
Overall best: Gemini 4 Argon (high) 76.0%
71st percentile · #6 of 18 open · #45 of 92 all
#6 of 18
#45 of 92
Independent
Epoch AI · Oct 2026

Other results (1)· leaderboard-only, not part of the llmrun Score

LiveBench 2026 Research
LiveBench 2026 Research measures finding and combining information to answer research questions. It belongs to the 2026 LiveBench release (first snapshot September 2, 2026), a redesigned LiveBench in which each of seven categories is scored by a single task and questions are refreshed to limit training-set contamination. It is a different test from the original LiveBench boards, so the scores are not comparable.
79.5
79.5
median 70.9open leader 86.8
Overall best: Claude Opus 5 88.9
88th percentile · #3 of 17 open · #18 of 53 all
#3 of 17
#18 of 53
Independent
LiveBench · Oct 2026

Score vs VRAM in the ≤512 GB tier

DeepSeek V4.1 Flash 1017
Other models
Best Score for its VRAM
x: VRAM (GB, log scale) · y: llmrun Score

#1 of 98 that fit ≤512 GB

Each dot is a model at the best quant that fits ≤512 GB. The line joins the models with the best Score for their VRAM; a dot below it is beaten by a model that needs no more memory.

Full ≤512 GB ranking →

DeepSeek V4.1 Flash 1017 vs its neighbours

Same boards for the nearest open models per llmrun Score. The highest score in each row is marked.

Board	DeepSeek V4.1 Flash 1017this model	MiMo V2.6 Pro RLnearest by ScoreQ4_K_M · 614 GB	GLM 5.3nearest by ScoreQ4_K_M · 454 GB	DeepSeek V4.1 Flashnearest by ScoreQ4_K_M · 458 GB
llmrun Score	133.2	133.2	130.1	129.4
GPQA Diamond	93.2%	—	90.9%	89.8%
Chess Puzzles	46.0%	—	37.0%	48.0%
LiveBench 2026 Reasoning	89.1	79.6	85.8	86.7
Mystery Game Puzzles	49.0%	—	33.0%	43.0%
LiveBench 2026 Coding	96.0	81.5	82.8	90.2
LiveBench 2026 Long-Horizon	50.9	47.5	35.9	40.1
AIME 2024/2025	99.4%	—	91.1%	98.3%
FrontierMath	71.2%	—	68.8%	67.4%
FrontierMath Tier 4	46.3%	—	29.3%	26.8%
SimpleQA	42.5%	—	41.0%	42.9%
Compare head to head		Compare	Compare	Compare

Scores aggregated from public benchmark sources (Epoch AI: 7 boards, LiveBench: 7 boards, Arena: 1 board) — each linked from the benchmark pages. llmrun does not run these benchmarks. Percentiles and ranks are among open models in the llmrun catalogue; the median and leader marks refer to the same field.