QBenchIndependent benchmark intelligence
DATA SNAPSHOT · 30 OCT 2026
Independent model evaluation · Public evidence

Model performance,
put into perspective.

Compare leading AI models across independent benchmark families. Inspect the underlying results, source dates and cost-performance trade-offs — without blending incompatible measurements.

QBENCH / EXTERNAL INDEX 1.0Last source snapshot: 30 Oct 2026
Comparable models
—
Exact model-name intersection
Benchmark families
—
Included in the composite
Source snapshots
—
Distinct benchmark datasets
Weight per family
20%
Equal-weight composite

Overall leaderboard

Equal-weight composite of five independent benchmark families · 0–100 source scores

COMPARABLE COHORT / 6 MODELS
Comparable model results
Six models with exact-name coverage across all five benchmark families
ModelLiveBenchGPQAFrontierMathProofBenchFurnitureComposite
Loading benchmark data…
Click a model row to open its profile. Search and provider filters only change the displayed rows; the composite cohort and scores remain unchanged.

Model analysis

Inspect one model or compare two models across the five composite inputs.

Benchmark profile
Published scores retained on their original 0–100 scale
—
Benchmark score matrix
Same six-model cohort · source values
FrontierMath column is the mean of its two published tiers. No missing scores are filled or estimated.

Visual analytics

Different views of the same published results — no synthetic or imputed scores.

VECTOR SVG / INTERACTIVE
Radar comparison
Five benchmark families · selected model and optional comparison
Parallel coordinates
Every comparable model across all five families
Score distribution
Five-family cohort spread · quartiles and median
Rank trajectory
How each model ranks within each benchmark family
Ranks are recalculated independently within each family from the six-model comparable cohort.
Cost vs accuracy frontier
GPQA Diamond · source-reported cost per question

Benchmark explorer

Open each source dataset and inspect model-level results, configuration and uncertainty where supplied.

8 DATASETS / 5 COMPOSITE FAMILIES + 2 SPECIALIST + 1 TIER SPLIT
Source dataset
Select a benchmark to inspect its published results.
Loading…

Included benchmark families

Sources used by the cross-benchmark composite. Coding specialist datasets remain separate.

Methodology & data policy

What the composite means — and what it does not mean.

TRANSPARENT / REPRODUCIBLE
01 / COHORT

Exact model identity

Only models present under the exact same model name in all five families enter the composite. No model aliases are inferred and no missing results are imputed.

02 / WEIGHTS

Equal family weighting

LiveBench 2.0, GPQA Diamond, FrontierMath, ProofBench and Furniture Assembly each contribute 20%. FrontierMath's two tiers are averaged first and count as one family.

03 / SOURCE SCALE

Published values retained

Source-reported 0–100 scores are retained as published. This is a descriptive composite, not a calibrated universal intelligence scale.

04 / SEPARATE DATA

Specialist results stay separate

SWE-bench Pro V2 and SWE-bench Verified are shown independently because their model coverage does not match the six-model composite cohort.

05 / AA EXCLUSION

No AA Index v5 inputs

None of the nine evaluations used by Artificial Analysis Intelligence Index v5 is included in this composite. QBench does not reproduce AA's scores, weighting or index.

06 / INTERPRETATION

Small gaps need context

Benchmarks differ in task design, difficulty, tool access, reasoning effort and harness. A small composite gap should not be read as a definitive model ranking for every use case.

Important: QBench's composite is an independent aggregation of public third-party benchmark snapshots. It is not a QBench-run evaluation, not an official ranking, and not a substitute for inspecting each benchmark's native result and methodology. Snapshot dates are shown beside every source.