Qwen3.6 27B — Benchmarks

Benchmark scores for Qwen3.6 27B aggregated from public leaderboards (11 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 →

111.2

90% range 107.1–115.2 · from 11 boards

Coding102.390% range 83–121
AgentsNot enough data yet
Math113.790% range 108–119
Science111.190% range 102–121
Reasoning105.190% range 98–113

Qwen3.6 27B 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.
85.9%
85.9%
median 55.7%open leader 93.2%
Overall best: Gemini 4 Argon (high) 96.4%
80th percentile · #21 of 100 open · #94 of 309 all
#21 of 100
#94 of 309
Independent
Epoch AI · Apr 2026
LiveBench Reasoning
LiveBench Reasoning measures logical, multi-step reasoning using contamination-free questions that are refreshed regularly, so models cannot have trained on the test set.
70.3
70.3
median 82.8open leader 90.7
Overall best: GPT 6 Astra Max 92.7
lowest · #20 of 20 open · #69 of 71 all
#20 of 20
#69 of 71
Independent
LiveBench · Jun 2026
CritPt
CritPt (Critical Points) tests graduate-level physics problem-solving on research-style questions spanning mechanics, electromagnetism, quantum mechanics, and more. It was created by physicists to probe whether a model can reason through multi-step physics rather than pattern-match textbook answers.
0.9%
0.9%
median 0.9%open leader 26.6%
Overall best: Gemini 4 Carbon (high) 36.0%
51st percentile · #43 of 84 open · #176 of 244 all
#43 of 84
#176 of 244
Lab-reported
Epoch AI · Apr 2026
DTBench
DTBench is a set of handcrafted, expert-validated multiple-choice questions on decision theory — Newcomb-like problems and related puzzles — built by the Conceptual Reasoning Index project with Anthropic. It tests whether a model can follow an abstract argument rather than recall a fact.
78.1%
78.1%
median 68.7%open leader 93.9%
Overall best: Claude Opus 5.5 Max 98.9%
62nd percentile · #27 of 70 open · #143 of 235 all
#27 of 70
#143 of 235
Lab-reported
Epoch AI · Apr 2026
LMCA
LMCA (Language Model Conceptual Argumentation) scores models on expert-rated arguments across philosophy, decision theory, and AI risk — domains where there is no empirical feedback loop to check an answer against. It is part of the Conceptual Reasoning Index, built with Anthropic by researchers Emery Cooper and Caspar Oesterheld.
34.5%
34.5%
median 25.5%open leader 55.5%
Overall best: Claude Sonnet 5.5 Max 68.3%
73rd percentile · #16 of 56 open · #129 of 197 all
#16 of 56
#129 of 197
Lab-reported
Epoch AI · Apr 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.
7.0%
7.0%
median 18.5%open leader 49.0%
Overall best: GPT 6 Astra Max 84.0%
11th percentile · #18 of 20 open · #125 of 136 all
#18 of 20
#125 of 136
Independent
Epoch AI · Apr 2026

Qwen3.6 27B coding benchmarks

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

LiveBench Coding
LiveBench Coding evaluates code generation and completion on fresh, contamination-free programming tasks that are updated regularly.
71.8
71.8
median 75.3open leader 81.5
Overall best: Claude Sonnet 5.5 Max 91.4
32nd percentile · #14 of 20 open · #59 of 71 all
#14 of 20
#59 of 71
Independent
LiveBench · Jun 2026
SciCode
SciCode asks a model to write real scientific-computing code — implementing multi-step numerical methods drawn from physics, chemistry, biology, and math research — and checks the output against reference solutions. It was built by a team of scientists and ML researchers to test coding ability on genuine research problems, not typical software tasks.
37.3%
37.3%
median 40.1%open leader 60.9%
Overall best: Claude Opus 5.5 Max 66.9%
37th percentile · #46 of 72 open · #186 of 217 all
#46 of 72
#186 of 217
Lab-reported
Epoch AI · Apr 2026

Qwen3.6 27B 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.
91.1%
91.1%
median 54.4%open leader 99.4%
Overall best: Claude Fable 5 (high) 100.0%
84th percentile · #14 of 88 open · #79 of 292 all
#14 of 88
#79 of 292
Independent
Epoch AI · Apr 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.
35.1%
35.1%
median 57.2%open leader 72.2%
Overall best: GPT 6 Astra Max 93.7%
17th percentile · #16 of 19 open · #69 of 117 all
#16 of 19
#69 of 117
Independent
Epoch AI · Apr 2026
LiveBench Math
LiveBench Math measures mathematical problem-solving on contamination-free, regularly-refreshed questions, including competition-style problems.
79.9
79.9
median 87.3open leader 95.1
Overall best: Claude Opus 5.5 Max 97.1
16th percentile · #17 of 20 open · #63 of 71 all
#17 of 20
#63 of 71
Independent
LiveBench · Jun 2026

Score vs VRAM in the 24 GB tier

Qwen3.6 27B
Other models
Best Score for its VRAM
x: VRAM (GB) · y: llmrun Score

#2 of 36 that fit 24 GB

Each dot is a model at the best quant that fits a 24 GB GPU. The line steps up to the best Score available at each amount of VRAM; a dot below it is beaten by a model that needs no more memory.

Full 24 GB ranking →

Qwen3.6 27B vs its neighbours

Same boards for the nearest open models per llmrun Score and the best model that fits 24 GB. The highest score in each row is marked.

Board	Qwen3.6 27Bthis model	Qwen3.8 27Bbest in classQ4_K_M · 18 GB	Qwen3.6 35B A3Bnearest by ScoreQ4_K_M · 22 GB	Gemma 4 31B ITnearest by ScoreQ4_K_M · 22 GB	Gemma 4 26B A4B ITnearest by ScoreQ4_K_M · 17 GB
llmrun Score	111.2	122.0	106.8	106.7	105.8
GPQA Diamond	85.9%	—	84.9%	75.8%	73.2%
LiveBench Reasoning	70.3	80.0	—	—	—
CritPt	0.9%	5.4%	0.3%	1.4%	0.0%
DTBench	78.1%	88.0%	73.9%	82.7%	74.9%
LMCA	34.5%	41.4%	29.7%	39.3%	29.7%
Mystery Game Puzzles	7.0%	—	22.0%	—	—
LiveBench Coding	71.8	75.7	—	—	—
SciCode	37.3%	46.6%	36.6%	45.5%	40.1%
AIME 2024/2025	91.1%	—	86.7%	73.3%	82.2%
FrontierMath	35.1%	—	20.3%	—	—
LiveBench Math	79.9	86.2	—	—	—
Compare head to head		Compare	Compare	Compare	Compare

Scores aggregated from public benchmark sources (Epoch AI: 8 boards, LiveBench: 3 boards) — 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.