Compare published benchmark results from matching versions and cohorts, API pricing, and technical specifications. Any documented test setup differences remain visible.
shared result series
191
distinct benchmarks
32
latest retrieval
2026-10-15
Change model selection
Specifications
General specifications
Release date, availability, context window, architecture, and published parameter counts at a glance. Undisclosed details remain marked as unavailable.
The table compares official standard API rates per 1M tokens, including published cache pricing. Batch, fast, priority, regional surcharges, tool calls, and cloud platform pricing are excluded.
Key to the marks:better valuebetter, but not conclusivetie
Pricing by provider
Direct developer pricing and provider endpoint tariffs routed through OpenRouter are listed separately. Prices are in USD per 1M tokens. Context-dependent direct tiers plus regional, Flex, Priority, and other OpenRouter tariffs remain individually identifiable.
No provider tariff for Muse Wonder is available in the stored OpenRouter snapshot. The official direct price is still shown.
Only published results with the same benchmark version, task, metric, and comparison cohort are paired. Different reasoning levels, agents, harnesses, or output limits appear directly in the table.
Capability profile from matched benchmarks
Each axis averages directly matched benchmark families. A value of 100 means the stronger value within this pair, not a universal quality score.
GPT-6 Astra
Muse Wonder
Coding: 98.6 / 95.3 (5 benchmark families)
Reasoning: 98.8 / 100 (1 benchmark family)
Knowledge: 97 / 100 (2 benchmark families)
Finance: 96.3 / 100 (2 benchmark families)
Legal: 86 / 100 (1 benchmark family)
Agentic tasks: 100 / 96.6 (3 benchmark families)
Productivity: 96.9 / 95 (2 benchmark families)
Health: 100 / 92.6 (1 benchmark family)
35 matched rows from 15 benchmark families were considered. The radar shows 8 of 10 comparable categories.
GPT-6 AstraMuse Wonder
AutomationBench-AA (strict pass@1)Overall. Higher is better. No documented setup difference
GPT-6 AstraWinner41.6 %
Muse Wonder39.1 %
BioMysteryBenchOverall. Higher is better. Test setup not documented as identical
GPT-6 AstraHigher measured value79.259 %
Muse Wonder67.037 %
Code MigrationOverall. Higher is better. Test setup not documented as identical
GPT-6 AstraHigher measured value67.74 %
Muse Wonder62.552 %
CyberBench v1.1Overall. Higher is better. Test setup not documented as identical
GPT-6 Astra41.072 %
Muse WonderHigher measured value80.357 %
DeepSWEOverall. Higher is better. No documented setup difference
GPT-6 Astra74.1 %
Muse WonderWinner78.8 %
EMBOverall. Higher is better. Test setup not documented as identical
GPT-6 AstraHigher measured value71.7 %
Muse Wonder67.084 %
Finance Agent (v2)Overall. Higher is better. Test setup not documented as identical
GPT-6 Astra53.54 %
Muse WonderHigher measured value56.637 %
GDP.pdfOverall. Higher is better. No documented setup difference
GPT-6 AstraWinner34.2 points
Muse Wonder30.8 points
Harvey LAB-AAOverall. Higher is better. No documented setup difference
GPT-6 Astra8.6 %
Muse WonderWinner10 %
Harvey's Legal Agent BenchmarkOverall · Task fully resolved. Higher is better. Test setup not documented as identical
GPT-6 Astra5.417 %
Muse WonderHigher measured value17.083 %
HealthBench ProOverall. Higher is better. No documented setup difference
GPT-6 AstraWinner65.3 points
Muse Wonder60.5 points
IOIOverall. Higher is better. Test setup not documented as identical
GPT-6 AstraHigher measured value100 %
Muse Wonder66.944 %
Key to the marks:better valuebetter, but not conclusivetie
No documented difference in the test setupVersion: Meta Wonder scorecard; version unspecifiedMetric: strict pass@1; every end-state assertion passedScoring: Higher is betterSettings: Reasoning: maxBusiness workflows in a simulated application environment. Every end-state assertion must pass; max reasoning. Do not merge with other AutomationBench-AA metrics.
Winner41.6 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
39.1 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
No documented difference in the test setupVersion: v1.1Metric: pass@1Scoring: Higher is betterSettings: Reasoning: max113 tasks across 91 repositories and five languages. Max reasoning. Meta attributes these results to Datacurve; the Muse scores could not be found on its public leaderboard at retrieval.
74.1 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
Winner78.8 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
No documented difference in the test setupVersion: Meta Wonder scorecard; version unspecifiedMetric: published GDP.pdf scoreScoring: Higher is betterSettings: Reasoning: maxProfessional work grounded in supplied PDFs. Max reasoning. Meta names Surge as the original source.
Winner34.2 points
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort. The source specifies neither a denominator nor a normalized scale. This is the published raw score, not a percentage.
30.8 points
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort. The source specifies neither a denominator nor a normalized scale. This is the published raw score, not a percentage.
No documented difference in the test setupVersion: 1.1Metric: all criteria passed with hallucination checksScoring: Higher is betterSettings: Reasoning: maxLegal deliverables must pass every criterion including hallucination checks. Max reasoning. Meta attributes the original evaluation to Artificial Analysis.
8.6 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
Winner10 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
No documented difference in the test setupVersion: Meta Wonder scorecard; version unspecifiedMetric: length-normalized weighted rubric scoreScoring: Higher is betterSettings: Reasoning: max525 health questions without tools; physician rubrics graded by GPT-5.4. Max reasoning. Meta attributes the Claude scores to Anthropic system cards.
Winner65.3 points
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort. The source specifies neither a denominator nor a normalized scale. This is the published raw score, not a percentage.
60.5 points
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort. The source specifies neither a denominator nor a normalized scale. This is the published raw score, not a percentage.
No documented difference in the test setupVersion: Meta Wonder scorecard; version unspecifiedMetric: mean task rewardScoring: Higher is betterSettings: Reasoning: max65 professional tasks across 35 occupations. Mean task-specific reward, max reasoning.
66.1 points
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort. The source specifies neither a denominator nor a normalized scale. This is the published raw score, not a percentage.
Winner70.5 points
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort. The source specifies neither a denominator nor a normalized scale. This is the published raw score, not a percentage.
No documented difference in the test setupVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterGPT-6 Astra: Snapshot: gpt-6-astra-max, Reasoning: max, Harness: Arena Code Arena WebDev Muse Wonder: Snapshot: muse-wonder-max, Reasoning: max, Harness: Arena Code Arena WebDevGPT-6 Astra: Arena rating from 5,746 published pairwise votes.Muse Wonder: Arena rating from 1,228 published pairwise votes.
Winner1,780.244
95% confidence interval: ±10.541The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,650.947
95% confidence interval: ±18.263The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
No documented difference in the test setupVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-6 Astra: Snapshot: gpt-6-astra-max, Reasoning: max, Harness: Arena Text, style control Muse Wonder: Snapshot: muse-wonder-max, Reasoning: max, Harness: Arena Text, style controlGPT-6 Astra: Arena rating from 11,674 published pairwise votes.Muse Wonder: Arena rating from 1,944 published pairwise votes.
1,476.716
95% confidence interval: ±6.385The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
Winner1,512.671
95% confidence interval: ±13.236The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
No documented difference in the test setupVersion: Meta internal Wonder evaluationMetric: LLM-judge weighted rubric scoreScoring: Higher is betterSettings: Reasoning: maxInternal Meta evaluation with 64K-100K-token conversations, distractors and conflicting or updated turns. Max reasoning.
81.4 points
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort. The source specifies neither a denominator nor a normalized scale. This is the published raw score, not a percentage.
Winner85.6 points
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort. The source specifies neither a denominator nor a normalized scale. This is the published raw score, not a percentage.
No documented difference in the test setupVersion: V2.1Metric: strict binary task completionScoring: Higher is betterSettings: Reasoning: max108 Ubuntu desktop workflows. Full task completion without partial credit, max reasoning. No binary Opus 5.5 score is published.
Tie34.3 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
Tie34.3 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
No documented difference in the test setupVersion: V2.1Metric: mean per-task partial rewardScoring: Higher is betterSettings: Reasoning: max108 Ubuntu desktop workflows. Partial reward, max reasoning. Meta attributes the Opus 5.5 score to Anthropic.
Winner75.8 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
72.6 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
Personal Superintelligence IndexOverallTest details
No documented difference in the test setupVersion: Meta Wonder launch scorecardMetric: Meta internal composite scoreScoring: Higher is betterSettings: Reasoning: maxMeta internal composite of seven personal and professional work evaluations. Max reasoning.
Winner54.5 points
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort. The source specifies neither a denominator nor a normalized scale. This is the published raw score, not a percentage.
54.3 points
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort. The source specifies neither a denominator nor a normalized scale. This is the published raw score, not a percentage.
No documented difference in the test setupVersion: Meta Wonder scorecard; version unspecifiedMetric: rollouts passing at least 95% of testsScoring: Higher is betterSettings: Reasoning: max, Harness: mini-swe-agentReconstruct programs from a specification. Common mini-swe-agent, clean environment without internet after setup. Share of rollouts passing at least 95% of tests, max reasoning.
Winner52.5 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
50 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
No documented difference in the test setupVersion: 4.0-AAMetric: task accuracyScoring: Higher is betterSettings: Reasoning: maxTerminal tasks in isolated environments with executable verifiers. Max reasoning. Meta names Artificial Analysis as the source; each agent remains part of the result.
Winner59.1 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
52 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
No documented difference in the test setupVersion: V2Metric: private test accuracyScoring: Higher is betterSettings: Reasoning: maxFinancial research using filings and market data. Private test set, max reasoning. Vals results republished by Meta.
53.5 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
Winner56.6 %
Comparison table published by Meta. Depending on the benchmark, Meta uses internal runs, provider reports or third-party results and selects the highest available validated score. This is not a uniformly independently measured cohort.
No documented difference in the test setupVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterGPT-6 Astra: Snapshot: gpt-6-astra-max, Reasoning: max, Harness: Arena Code Arena WebDev Muse Wonder: Snapshot: muse-wonder-max, Reasoning: max, Harness: Arena Code Arena WebDevGPT-6 Astra: Arena rating from 5,746 published pairwise votes.Muse Wonder: Arena rating from 1,228 published pairwise votes.
Winner1,780.244
95% confidence interval: ±10.541The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,650.947
95% confidence interval: ±18.263The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
LMArena Text, Style ControlBusiness, management and financeTest details
No documented difference in the test setupNo clear leadVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-6 Astra: Snapshot: gpt-6-astra-max, Reasoning: max, Harness: Arena Text, style control Muse Wonder: Snapshot: muse-wonder-max, Reasoning: max, Harness: Arena Text, style controlGPT-6 Astra: Arena rating from 2,061 published pairwise votes.Muse Wonder: Arena rating from 338 published pairwise votes.
1,472.34
95% confidence interval: ±13.336The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
No clear lead1,500.247
95% confidence interval: ±30.856The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
No documented difference in the test setupNo clear leadVersion: webdevMetric: WebDev Arena Bradley-Terry ratingScoring: Higher is betterGPT-6 Astra: Snapshot: gpt-6-astra-max, Reasoning: max, Harness: Arena Code Arena WebDev Muse Wonder: Snapshot: muse-wonder-max, Reasoning: max, Harness: Arena Code Arena WebDevGPT-6 Astra: Arena rating from 1,129 published pairwise votes.Muse Wonder: Arena rating from 158 published pairwise votes.
No clear lead1,780.13
95% confidence interval: ±22.887The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
1,721.908
95% confidence interval: ±51.47The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
No documented difference in the test setupNo clear leadVersion: style-control-v1Metric: style-controlled Bradley-Terry ratingScoring: Higher is betterGPT-6 Astra: Snapshot: gpt-6-astra-max, Reasoning: max, Harness: Arena Text, style control Muse Wonder: Snapshot: muse-wonder-max, Reasoning: max, Harness: Arena Text, style controlGPT-6 Astra: Arena rating from 2,659 published pairwise votes.Muse Wonder: Arena rating from 487 published pairwise votes.
1,543.161
95% confidence interval: ±11.756The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
No clear lead1,557.956
95% confidence interval: ±26.189The Arena score measures human preference rather than a fixed capability test. The confidence interval, Arena category, and model configuration are part of the result.
Your workload, budget, and required context length matter most. A coding benchmark says little about visual understanding or agent performance.
The standard price table uses direct developer rates. The provider comparison labels OpenRouter endpoints, regions, and special tariffs separately.
Every result links to its measurement source and retrieval date. Labels distinguish vendor reports, official benchmark leaderboards, and independent evaluations. Disputed or archived results are excluded.
More matchups
Related LLM comparisons
Compare GPT-6 Astra and Muse Wonder with other leading models using the same data and benchmark logic.
