UnifyBench

RankingEffort benchmarks

Qwen3-4B effort benchmarks

Compare measured Low, Medium, High and other settings. Every number keeps its source.

Qwen3-4B

17 measured results · 3 reported effort labels · 11 benchmark/harness combinations. Model evidence and sources →

Capability score evidence

Qwen3-4B · Reasoning: 1.4 overall comparison score. Meets the overall evidence requirements.

Which sources carry the weight?
  • Artificial Analysis · 100.0% direct comparison weight · independent

These shares describe the comparisons entering the fit, not fractions of the final score.

How sensitive is this score?

Remove one evaluator or family and refit the graph against the same overall panel. Values use the matching dataset audited 2026-09-12. These are sensitivity checks, not confidence intervals; ranking eligibility may also change.

Removed evidenceRecomputed scoreChange
Artificial AnalysisComparison path lost—
ARC Prize1.7+0.3
Datacurve1.1-0.2
LiveCodeBench1.3-0.0
Scale1.5+0.2
Terminal-Bench1.3-0.0
Cognition1.4+0.1
LiveBench1.4+0.0
SWE-rebench1.4+0.0
Epoch AI1.4+0.0
Vals AI1.3-0.0
Provider reports · Anthropic1.4+0.0
Provider reports · Moonshot AI1.4-0.0
Provider reports · Meta1.3-0.0
Provider reports · Mistral1.4+0.0
All provider reports1.3-0.0
Every benchmark family
Removed evidenceRecomputed scoreChange
arc-agi1.6+0.3
deepswe1.1-0.2
math5001.2-0.1
aa-lcr1.5+0.1
mmmu-pro1.5+0.1
livecodebench1.3-0.1
omniscience1.3-0.1
aime1.4+0.1
hle1.4+0.1
frontiercode1.4+0.1
epoch-game-puzzles1.3-0.1
simpleqa1.4+0.1
mmlu-pro1.4+0.1
critpt1.3-0.1
swe-rebench1.4+0.0
ifbench1.3-0.0
gpqa1.3-0.0
scicode1.4+0.0
multi-swe-bench1.3-0.0
livebench-instructions1.4+0.0
gdp-pdf1.4+0.0
livebench-language1.4+0.0
apex-agents1.4+0.0
gmmlu1.4+0.0
harvey1.3-0.0
gdpval1.4+0.0
frontiermath1.4+0.0
swe-atlas-qna1.3-0.0
terminal-bench1.4+0.0
livebench-coding1.3-0.0
vibe-code1.3-0.0
livebench-reasoning1.3-0.0
arc-agi-31.3-0.0
vals-finance-agent1.3-0.0
vals-legal-research1.3-0.0
analyst-agent1.4+0.0
swe-atlas-test-writing1.3-0.0
vals-code-migration1.3-0.0
enterprise-ops1.4+0.0
tau-bench1.3-0.0
livebench-data1.4+0.0
vals-excel-modeling1.3-0.0
livebench-math1.4+0.0
swe-atlas-refactoring1.4-0.0
briefcase1.4-0.0
enigma-eval1.4+0.0
itbench1.4-0.0
automationbench1.4+0.0
swe-bench-pro1.4+0.0
What evidence is missing from the fit?

Across all collected settings for this model: 17 observations, 16 contributing, 0 matched without graph weight and 1 excluded. Counts do not establish rank eligibility.

  • 1 · Effort is not identified
Inspect every observation and exclusion →

Capability profile

One configuration, seven capabilities

Comparison score · 0–100
Qwen3-4B · Reasoning capability radarComparison scores from 0 to 100. Unknown capabilities have no point; lines stop at gaps. Hollow points are preliminary. Exact values and support labels follow the chart.AgenticHard reasoningCodingHuman prefKnowledgeMultimodalLong context50100Qwen3-4B · Reasoning · Agentic: 0.9 · PreliminaryQwen3-4B · Reasoning · Hard reasoning: 1.9 · SupportedQwen3-4B · Reasoning · Coding: 2.2 · PreliminaryQwen3-4B · Reasoning · Knowledge: 1.7 · SupportedQwen3-4B · Reasoning · Long context: 0.8 · Preliminary

Gaps are unknown, not zero. Hollow points are preliminary.

  • Agentic

    A0.9Preliminary

  • Hard reasoning

    A1.9Supported

  • Coding

    A2.2Preliminary

  • Human pref

    Unknown

  • Knowledge

    A1.7Supported

  • Multimodal

    Unknown

  • Long context

    A0.8Preliminary

Each profile belongs to the named configuration. Capability scores use a shared panel for all axes, distinct from the overall ranking panel; they are not accuracy percentages or direct head-to-head wins.

Compare this configuration →Explore the model evidence profile →

These are published benchmark scores. Supported effort estimates also appear on the Capability leaderboard. Results from different harnesses occupy different rows. Missing cells stay unknown; no scores are copied between effort levels.

Benchmark / sourceReasoningNon-reasoningUnspecified
AIME 2025Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

22.33%
Reported settings

Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning

Not reportedNot reported
AA-LCR v1.1Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

0%
Reported settings

Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning

Not reportedNot reported
GPQA DiamondArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

52.22%
Reported settings

Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning

39.8%
Reported settings

Qwen3 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct

Not reported
Humanity's Last ExamArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

4.42%
Reported settings

Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning

3.32%
Reported settings

Qwen3 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct

Not reported
IFBenchArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

32.52%
Reported settings

Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning

Not reportedNot reported
LiveCodeBenchArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

46.46%
Reported settings

Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning

23.28%
Reported settings

Qwen3 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct

Not reported
MATH-500Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

93.33%
Reported settings

Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning

84.27%
Reported settings

Qwen3 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct

Not reported
MMLU-ProArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

69.56%
Reported settings

Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning

58.59%
Reported settings

Qwen3 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct

Not reported
𝜏²-Bench TelecomArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-09-12.

19.01%
Reported settings

Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning

Not reportedNot reported
Epoch · Chess Puzzles · v1.1.6Epoch AI · Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match
Benchmark details

Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.

Unit: percent. Source collected 2026-09-12.

Not reported
1%
Reported settings

Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3-4b_none; run=3mCgC7byPaZ3zzPM98bY5A; mean accuracy; https://epoch.ai/data/benchmarks.csv

Not reported
Epoch · GPQA diamond · v1.0.11Epoch AI · Epoch AI Inspect; task 1.0.11; scorer choice
Benchmark details

Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.

Unit: percent. Source collected 2026-09-12.

Not reported
43.56%
Reported settings

Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-4b_none; run=SLJyD8yew4TxcmprJfVvuE; mean accuracy; https://epoch.ai/data/benchmarks.csv

52.34%
Reported settings

Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-4b; run=G3dinPvwrci2Nu3m9pby2i; mean accuracy; https://epoch.ai/data/benchmarks.csv

What these comparisons mean

Each source reports its own effort labels. Matching labels do not establish equal compute budgets or identical fallback behavior. Open Reported settings to inspect the exact model name, agent and task coverage. Multiple results at one level remain visible rather than selecting the best. Source-reported None or Non-reasoning labels are not treated as evidence that an API supports those settings.

These measurements use their own dated source collection. The leaderboard fits identified configurations jointly; it contains no pooled model ratings. Explicit thinking budgets are separate configurations. Read the methodology.

Download all effort benchmark results · 7,424 results across 240 models · Collected 2026-09-12