‹ PublicAI Index
The LLM benchmark aggregator.
Qwen3.8 Max
Alibaba
Strongest in Science (#6 of 122), weakest in Toxicity avoidance (#184 of 268). Above par in 27 of 32 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5.1 and Claude Fable 5 and ahead of MiMo V2.6 Pro and GLM-5.3.
§ 1 · Profile
What it is good at
Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.
Human preference64.7−4.1#20/3541/1
Human preference64.7−4.1#20/3541/1
Knowledge53.8−8.7#48/1971/3
Academic knowledge59.2−4.7#18/1231/1
Reasoning54.6−10.8#52/1953/5
Science62.3−1.2#6/1221/1
Reasoning53.7−15.3#50/1471/3
Mathematics51.6−12.2#76/1522/2
Professional53.5−8.9#54/1942/2
Document design55.5−13.2#10/311/1
Finance57.9−4.5#18/1621/1
Legal57.1−8.8#27/1731/1
Cybersecurity26.3−37.6#47/471/1
Medical51.9−14.6#75/1491/1
Core abilities54.3−11#54/1993/4
Instruction following60.8−13#13/621/1
Data analysis56.4−8.8#14/311/1
Language52.1−14.5#31/621/1
General intelligence51.1−18.3#78/1782/3
Coding53.2−13#56/1794/6
Agentic coding53.3−13.6#55/1734/5
Agents53.2−13.2#78/2983/6
Knowledge work66.5−4.2#12/2082/2
Long-horizon work44.1−29.9#17/311/1
Web research45.5−18.4#27/351/2
Computer use46.9−21.2#42/681/2
Safety53.1−8.3#97/3291/3
Secure code63.6−2.6#20/2701/1
Harm refusal52.5−9#117/2921/2
Fairness48.4−21.6#134/2921/2
Jailbreak resistance53.2−11.8#140/2681/1
Toxicity avoidance51−8#184/2681/1
§ 2 · Head to head
What it beats, and what beats it
The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Qwen3.8 Max, left for the other.
§ 3 · Sources
Where the numbers come from
9 publications, 43 figures. Every one links to the page it was read from.
LMArena Text 1484
Artificial Analysis Intelligence Index 34
GDPval-AA 1674
AA-Briefcase 1605
Terminal-Bench 27% (Claude Code)
LiveBench 60.9LiveBench · Agentic 26.6LiveBench · Reasoning 88.2LiveBench · Coding 60.9LiveBench · Browser Use 49.4LiveBench · Data Analysis 78.4LiveBench · Research 72.8LiveBench · Design 76.6
LiveBench 1.0 78.5LiveBench 1.0 · Reasoning 88.2LiveBench 1.0 · Coding 72.9LiveBench 1.0 · Agentic Coding 64.6LiveBench 1.0 · Mathematics 91.3LiveBench 1.0 · Data Analysis 78.4LiveBench 1.0 · Language 79.7LiveBench 1.0 · Instruction Following 74.1
Vals · Legal Research Bench 46.15%Vals · LegalBench 83.61%Vals · Harvey Legal Agent Benchmark 15.42%Vals · Finance Agent 57.87%Vals · CorpFin 65.85%Vals · TaxEval 75.55%Vals · MortgageTax 63.99%Vals · MedCode 40.67%Vals · MedScribe 84.95%Vals · CyberBench 28.57%Vals · SWE-bench Verified 85.6% (Mini-SWE-agent)Vals · Vibe Code Bench 77.14% (OpenHands)Vals · Code Migration 23.96%Vals · GPQA Diamond 93.69%Vals · MMLU Pro 88.6%Vals · ProofBench 58%
Enkrypt · Jailbreak risk 10.1%Enkrypt · Harmful content risk 0.6%Enkrypt · CBRN risk 26%Enkrypt · Toxicity risk 5.5%Enkrypt · Bias risk 80.4%Enkrypt · Insecure code risk 5.3%
Badge
[](https://publicai.io/model-index/m/qwen3-8-max)