Qwen 3.8 27B

Qwen 3.8 27B places in the top three on 4 benchmarks.

Qwen Released August 14th, 2026 Compare with others

108

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Qwen Aug 2026

Qwen 3.8 27B

Open weights

152.6 #58 of 256

Epoch Capabilities Index

Self-Reported Index 153.3 #73 of 404

IFBench 79.5% #11 /80
NL2Repo-Bench 42.3% #21 /30
Agents' Last Exam 20.4% #21 /31

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
BullshitBench	76.4%	#8 of 107	Official leaderboard
AgentIF (2026-10 setting)	65.8%	#2 of 24	Official leaderboard
BullshitBench V2	79%	#13 of 129	Official leaderboard
IFBench	79.5%	#11 of 80	Lab model card
LMArena (Elo)	1,433	#88 of 335	Official leaderboard
Judgemark v4	67.4	#17 of 40	Official leaderboard
Humanity's Last Exam	30.8%	#78 of 154	Lab model card
Creative Writing v3	64.8	#71 of 120	Official leaderboard
Lech Mazur Creative Writing	-0.70	#31 of 50	Official leaderboard
Longform Creative Writing (EQ-Bench)	53.3	#79 of 119	Official leaderboard
Extended NYT Connections	54.5%	#69 of 103	Official leaderboard
StudyArena Text	1,219	#57 of 85	Official leaderboard
QIMMA	42.3%	#21 of 25	Official leaderboard

Math and science

Benchmark	Score	Rank	Source
GPQA Diamond	89.2%	#42 of 302	Lab model card
PHYSICS-Verified	87.2%	#4 of 19	Official leaderboard
DTBench	88%	#59 of 180	Official leaderboard
SimpleBench	60.2%	#43 of 106	Official leaderboard
ARC-AGI	87.5%	#41 of 95	Official leaderboard
ARC-AGI-2	42.4%	#45 of 98	Official leaderboard
Surface Evolver Bench	45%	#18 of 25	Official leaderboard
ProofBench v1.1	16%	#46 of 52	Official leaderboard

Coding

Benchmark	Score	Rank	Source
LiveCodeBench v6	90.3%	#6 of 90	Lab model card
WebDev Arena	1,593	#20 of 105	arena.ai, via Epoch AI
APEX-SWE	50.7%	#14 of 65	Official leaderboard
Code Arena	1,590	#34 of 128	Official leaderboard
SWE-Bench Pro	61.7%	#25 of 81	Lab model card
Image-to-WebDev Arena	1,574	#23 of 55	Official leaderboard
Vibe Code Bench	64.8%	#55 of 120	Official leaderboard
SWE-bench Multimodal	38.6%	#8 of 12	Lab model card
NL2Repo-Bench	42.3%	#21 of 30	Lab model card
SWE-bench Science	29.4%	#7 of 10	Official leaderboard
Vision2Web	62.9%	#5 of 7	Lab model card
Code Migration	14.2%	#58 of 78	Official leaderboard
IOI v2 (Vals)	39.1%	#37 of 45	Official leaderboard
DeepSWE	42.2%	#52 of 56	Lab model card
QwenSWEBench	79%	#2 of 2	Lab model card

Agents and tools

Benchmark	Score	Rank	Source
OSWorld-Verified	84.3%	#4 of 41	Lab model card
AndroidWorld	81.9%	#4 of 23	Lab model card
GameWorld	54.6%	#4 of 22	Official leaderboard
ScienceBoard	42%	#3 of 15	Official leaderboard
LLM Chess	846	#28 of 123	Official leaderboard
CivBench	1,516	#7 of 19	Official leaderboard
GameHorizon	68.5	#16 of 37	Official leaderboard
WildClawBench	48%	#16 of 31	Official leaderboard
Gym-Anything-RT	1.00%	#3 of 5	Official leaderboard
Terminal-Bench 2.1	73%	#35 of 56	Lab model card
AdaExploreBench	50.6%	#5 of 8	Official leaderboard
o11y-bench	64.1%	#15 of 22	Official leaderboard
WebArena-Verified	64.8%	#5 of 7	Lab model card
Agent Arena	-4.76%	#42 of 56	Official leaderboard
WitnessBench	3.10%	#10 of 12	Official leaderboard
RuneBench	2.79	#68 of 76	Official leaderboard
DuelLab GameBench 2	16.6	#51 of 57	Official leaderboard
BabaIsBench	21%	#32 of 32	Official leaderboard
RecreationBench	47.1%	#3 of 3	Lab model card

Professional work

Benchmark	Score	Rank	Source
BRIDGE	55.7	#2 of 39	Official leaderboard
Harvey LAB (Legal Agent Benchmark)	11.3%	#15 of 80	Official leaderboard
Tax Agent Bench	30.9%	#32 of 84	Official leaderboard
MedScribe	83.8%	#39 of 97	Official leaderboard
FoodTruck Bench	$27,168	#26 of 59	Official leaderboard
Excel Modeling Benchmark	59.7%	#36 of 76	Official leaderboard
MortgageTax	64.9%	#42 of 87	Official leaderboard
LegalBench	82.4%	#70 of 138	Official leaderboard
Legal Research Bench	36.1%	#49 of 93	Official leaderboard
Vals Index v2	48.5%	#37 of 65	Official leaderboard
APEX-1	62.6%	#41 of 71	Official leaderboard
Vals Finance Agent v2	48.5%	#49 of 80	Official leaderboard
TaxEval v2	70.8%	#80 of 129	Official leaderboard
APEX-Agents	47.5%	#38 of 61	Collected by Epoch AI, source not named
APEX-Agents 1.1	62.2%	#44 of 65	Official leaderboard
Agents' Last Exam	20.4%	#21 of 31	Official leaderboard
TRLawBench	8.46	#24 of 34	Official leaderboard
CoWorkBench	70.7%	#5 of 7	Lab model card
BlenderWorld	47.7	#10 of 13	Official leaderboard
Investor's Last Exam	54.6	#9 of 10	Official leaderboard
MedCode	28.7%	#91 of 96	Official leaderboard
JobBench	33.4%	#15 of 15	Lab model card

Vision, audio and robotics

Benchmark	Score	Rank	Source
CharXiv	90.2%	#7 of 75	Lab model card
MATH-Vision (MathVision)	94.6%	#5 of 51	Lab model card
RealWorldQA	85.9%	#7 of 38	Lab model card
BabyVision	85.6%	#8 of 26	Lab model card
ParseBench	69%	#15 of 44	Official leaderboard
ERQA	65.5%	#13 of 36	Lab model card
Vision Arena	1,243	#51 of 136	Official leaderboard
ExtractBench	89.8%	#11 of 27	Official leaderboard
OCRBench v2	63.5%	#6 of 14	Official leaderboard
Claw-Eval-MM	56.9%	#5 of 6	Lab model card
Code2Video-Compass	13.8	#14 of 15	Official leaderboard
GeoBench	38%	#45 of 48	Official leaderboard

Safety and security

Benchmark	Score	Rank	Source
BioTIER Permit	99.8%	#17 of 69	Official leaderboard
Molecular Biology Capabilities Test	53%	#27 of 90	Official leaderboard
ABC Bench Fragment Design	90.7%	#13 of 27	Official leaderboard
ABC Bench Liquid-Handling Robot	90.3%	#19 of 34	Official leaderboard
CyberBench v1.1 (Vals)	67.8%	#25 of 44	Official leaderboard
Human Pathogen Capabilities Test	49%	#51 of 86	Official leaderboard
nocot-bench (NCRI)	66.1	#116 of 185	Official leaderboard
World-Class Bio	40%	#48 of 74	Official leaderboard
Human Pathogen Capabilities Test v2 (HPCT v2)	48.6%	#47 of 71	Official leaderboard
ABC Bench Screening Evasion	53.9%	#13 of 17	Official leaderboard
BioTIER Refuse	21.9%	#55 of 69	Official leaderboard
VCT-v2	28%	#23 of 25	Official leaderboard

Other

Benchmark	Score	Rank	Source
SAGE (Vals)	52.4%	#7 of 81	Official leaderboard
ACCoRD	73.3	#35 of 152	Official leaderboard
Conceptual Reasoning Index	55.5	#53 of 151	Official leaderboard
LMCA	41.4%	#53 of 147	Official leaderboard
Decision Index	57.0	#2 of 3	Official leaderboard
LiveBench	54%	#26 of 31	Official leaderboard
Political Science LLM Benchmark	63.5%	#36 of 42	Official leaderboard

Its lab's numbers put it at 153.3, in line with its ECI

The Self-Reported Index fits one score to the 17 capability results Qwen reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 150.3 to 156.3. How we compute it.

Its lab's number next to a leaderboard

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark	Lab says	Board says	Gap
Agents' Last Exam	20.4% Frontier agentic tasks Agents' Last Exam Pass@1 20.4 Score 42.9	20.4% Leaderboard, max, claude_code	Same

Qwen's models over time

Qwen's models in color, against every other model on the Epoch Capabilities Index, by release date.

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Qwen3.8-27B model card (Hugging Face) Aug 14, 2026