Atlas

Compare Models

Opus 4.8 vs Qwen3.8-Max

Opus 4.8 leads by 0.1 capability points (≈95% interval −2.2 to +2.3): statistically tied · P(higher) ≈ 52%

Qwen3.8-Max wins 53 of 79 shared benchmarks (26 lost)

Capability vs. relative cost
Opus 4.8Qwen3.8-MaxClosed weightsOpen weights

Each line runs through one model’s priced reasoning settings, cheapest to most expensive; grey points are other models’ priced settings in the same range. Results whose reasoning effort the source didn’t report are left out wherever a model has named effort levels. Capability and relative cost are the published estimates (relative cost is cost per task as a multiple of GPT-5 (high)); hover a setting, or use the arrow keys, to set it beside the other model’s same setting.

Opus 4.8vsQwen3.8-Maxcapability vs. relative costData Nov 15, 2026Village LLM Atlasatlas.village.dev

Benchmark wins26 – 53 · 79 shared

Comparing matching setups only (174 benchmarks excluded) · suites counted once
← Opus 4.8Qwen3.8-Max →

Opus 4.8vsQwen3.8-Maxwins by categoryData Nov 15, 2026Data: Alibaba, Artificial Analysis, Conceptual Reasoning Benchmark Authors and 16 othersVillage LLM Atlasatlas.village.dev

Margins

Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Qwen3.8-Max minus Opus 4.8.

BabyVisionMultimodal28.482.0+53.6
BabaIsBench Stage 0: The IntroGames96.058.0−38.0
MLVU (multiple-choice M-Avg)Multimodal53.490.8+37.4
Medical Long Context Reasoning (MLCR-AA)Health & Medicine45.619.4−26.1
ERQAMultimodal57.277.8+20.6
IFBench - release-reported instruction followingGeneral QA62.282.8+20.6
WorldVQA AccuracyMultimodal33.953.2+19.3
LMArena Code Arena WebDevCode15361670+17.6
MRCR v2 (8-needle, Context Arena) - 128K pointwiseGeneral QA75.292.3+17.2
𝜏³-BankingAgents34.251.3+17.1
PerceptionBenchMultimodal47.263.5+16.3
MMVUMultimodal67.482.4+15.0
ZeroBench (Main questions, pass@5, with tools)Multimodal34.049.0+15.0
SLAKEHealth & Medicine75.990.8+14.9
LVBenchMultimodal67.381.8+14.5
NL2Repo - release-reported repository generation scoreCode69.755.9−13.8
YC-BenchGames17113892095838+13.4
Video-MMMUMultimodal75.388.7+13.4
AA-Omniscience IndexIndexes28.83.4−12.7
Conceptual Reasoning IndexIndexes68.359.4−12.3
RealWorldQA - release-reported visual perceptionMultimodal76.688.0+11.4
LMCAGeneral QA57.546.2−11.3
GameBench 2Games62.551.4−11.0
LingoQAMultimodal73.884.8+11.0
AndroidWorld - release-reported mobile useAgents75.085.3+10.3
CharXiv ReasoningMultimodal78.588.4+9.9
DeepsecBenchCode6.916.5+9.6
MMStarMultimodal76.785.9+9.2
MedXpertQA MMHealth & Medicine71.780.4+8.7
DeepsecBench — Revalidated V1Code6.915.5+8.6
DeepsecBench — Revalidated V1 (233 issues)Code6.915.5+8.6
MRCR v2 (8-needle, Context Arena) - 128K AUCGeneral QA87.896.1+8.3
MathVision - Qwen corrected annotationsMath87.195.2+8.1
OSWorld 2.0 - reported partial reward, patch and step limit unspecifiedAgents54.846.7−8.1
HealthBenchHealth & Medicine52.460.2+7.8
SimpleVQAMultimodal67.375.0+7.7
Agents' Last Exam - release-reported score, subset unspecifiedAgents45.152.4+7.3
PMC-VQAHealth & Medicine59.266.2+7.0
MMMU ProMultimodal75.682.3+6.7
Ruby LLM Benchmarks — Program FixerCode67.474.0+6.6
GDPval-AA v2.1Professional Work14611611+6.6
Vision2Web - category-average web development scoreCode62.469.0+6.6
ACCoRDGeneral QA77.471.4−5.9
OmniDocBench v1.5Multimodal86.592.1+5.6
SCI-AGENT-EXAMScience57.162.7+5.6
Vision Arena OverallMultimodal12791299+5.2
JobBench - release-reported aggregation unspecifiedProfessional Work48.453.4+5.0
Video-MME (with subtitles)Multimodal85.490.4+5.0
GDPval-AA v2Professional Work15291630+5.0
GPQA DiamondGeneral QA90.994.9+4.0
ClawEval-MM - pass@3Agents73.377.2+3.9
RecreationBench - Qwen internal application recreationCode48.051.7+3.7
Toolathlon-VerifiedAgents76.272.5−3.7
AutomationBench-AA 1.0.6Agents45.649.2+3.6
Finance Agent v2Finance & Accounting53.950.6−3.3
Terminal-Bench 4.0 (Artificial Analysis)Code21.718.7−3.0
xbench-ScienceQAScience76.473.4−3.0
DTBenchGeneral QA94.992.0−2.9
LongBench v2General QA69.166.3−2.8
OSWorld-VerifiedAgents83.486.1+2.7
GDP.pdf (Artificial Analysis, All-pass)Professional Work22.820.2−2.6
CoWorkBench - Qwen internal professional workProfessional Work72.374.8+2.5
ScreenSpot-ProMultimodal82.384.5+2.2
BuseyBench SVGMultimodal5.25.4+2.1
Epoch Capabilities Index (ECI)Indexes160157−2.0
AA-Briefcase EloAgents13541388+2.0
Humanity's Last Exam (Text-Only)General QA44.843.0−1.8
Humanity's Last Exam (with tools)General QA57.956.2−1.7
LMArena Text Arena EloChat & Writing14741484+1.7
Artificial Analysis Intelligence Index v4.3Indexes42.040.3−1.7
Artificial Analysis Intelligence Index v4.3.2Indexes41.840.2−1.6
SWE-bench Pro - Qwen corrected task cohortCode69.267.7−1.5
DeepSWE v1.1Code58.056.6−1.4
OSWorld 2.0 - reported binary completion, patch and step limit unspecifiedAgents20.619.4−1.2
SciCode 1.0.1 (Artificial Analysis)Code54.453.2−1.2
ClawEval-MM - average of three attemptsAgents73.874.8+1.0
Terminal-Bench 2.1Code82.081.3−0.7
Artificial Analysis Progress Index v1Indexes318320+0.4
AutomationBench (Public Set)Agents27.227.3+0.1

Opus 4.8vsQwen3.8-Maxnormalised score marginsData Nov 15, 2026Data: Alibaba, Artificial Analysis, Conceptual Reasoning Benchmark Authors and 16 othersVillage LLM Atlasatlas.village.dev

Win counts use each benchmark’s declared score direction. Represented suite slices and the derived Capability score are excluded from headline totals. Missing or incompatible setups are excluded, never counted as losses.
Loading Atlas data…