Muse Wonder: score and benchmark evidence
Historical release · · Meta
127.0 points · Rank 9 among 37 ranked models · Stability range 114.9–138.4 points.
Rank follows the point estimate; it does not establish statistically significant superiority. The scale is anchored at mean 100, SD 15 in the frozen calibration cohort, not human IQ or a percentage.
One score summarizes demonstrated capability across published reasoning settings and seven equally weighted domains. Related benchmark variants share a family budget, and influence from one evaluator is capped.
Release tli-2026-v1.0-2026-10-26-kimi-k3-1-pro-lab-6. Methodology and limitations for this release.
This page preserves historical evidence. See the current model record.
Capability domains
Domain scores use calibration-cohort standard-deviation units. Unmeasured domains are shown as unavailable, not zero.
| Domain | Score (SD units) | Observed benchmarks |
|---|---|---|
| Knowledge & reasoning | 0.58 | 5 |
| Coding & software engineering | 0.97 | 5 |
| Agentic tool & computer use | 0.78 | 2 |
| Professional & real-world work | 0.88 | 8 |
| Multimodal & vision | Unavailable | 0 |
| Cybersecurity | Unavailable | 0 |
| Preference & communication | Unavailable | 0 |
Published benchmark evidence
20 results contribute to this score, from 21 collected results and 20 contributing benchmark families. Counts are not independent sample sizes.
Results, source citations, review decisions and compatible reasoning modes are stored in a versioned relational database. Each publication uses a sealed, checksum-verified export; CSV files are optional exports, not editable scoring inputs. Previous revisions remain available for audit. Collected means a numeric result exists in the database. Used means its source, exact model checkpoint and benchmark protocol have been reviewed and its benchmark has a frozen calibration supported by multiple providers. For each benchmark we select the highest published aggregate across comparable reasoning modes, not necessarily maximum effort, and count it once. More tested settings provide more chances to record a high score; this selection advantage is not corrected by the conditional interval. Developer-reported and sole/unspecified settings remain eligible and disclosed. Unresolved does not mean false or useless: conflicting metrics, unverified citations and unmatched versions stay collected until reconciled.
The score stability range is a 95% bootstrap score interval conditional on frozen calibration and comparable evidence availability. It resamples benchmark families and evaluators and adds a frozen missing-evidence correction. It excludes selection across reasoning settings, selective publication, calibration changes and new domains. It is not a 95% guarantee about true capability or where a future score will land. Sparse or conflicting evidence can widen the range.
Reporting sensitivity: ±10.5 points. Sensitivity to selectively published results. This stress range is not a confidence interval or a probability statement.
AA-Briefcase v1.1 (Elo): 1658.65 native-points
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 1729.32 recorded 2026-10-12.
GDPval-AA v2.1 (Elo): 1770.34 native-points
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 1857.09 recorded 2026-10-12.
AutomationBench-AA: 68.094551%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 71.3307497322 recorded 2026-10-12.
Terminal-Bench 4.0 — AA harness: 52.020202%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 57.0707070707 recorded 2026-10-12.
SciCode: 60.300926%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 62.6157407407 recorded 2026-10-12.
HLE text-only — AA harness: 50.092678%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 53.6144578313 recorded 2026-10-12.
GDP.pdf — AA harness: 30.6%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 32.6 recorded 2026-10-12.
CritPt: 28.285714%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 32.8571428571 recorded 2026-10-12.
AA-Omniscience: 25.483333 native-points
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 30.15 recorded 2026-10-12.
AA-LCR v1.1: 80.666667%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 83.6666666667 recorded 2026-10-12.
AA Intelligence Index v4.3.2 (score): 52.393895 native-points
Not used in the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Artificial Analysis independent evaluation, 2026-10-15; fallback not reported
Metric: published native-points
Original source · Retrieved 2026-10-15
Daily evaluator refresh; replaces 55.8021199856 recorded 2026-10-12.
DeepSWE v1.1: 78.8%
Contributes to the score. Developer-reported. Review status: reviewed-developer.
Configuration: Muse Wonder (max); Muse Wonder developer-reported result (meta); fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-12
Developer-reported result entered in the newsroom. Same protocol as the catalogue row: the lab table reproduces cells we hold (opus-5-5 74.2, fable-5-1 67.4). Value read from an image of the lab table. Meta's methodology says the DeepSWE scorecard uses Datacurve's third-party results for Muse Wonder and Claude Opus 5.5, OpenAI's provider-reported result for GPT-6 Astra, and Anthropic's system card for Claude Fable 5.1. Metric pass@1.
Finance Agent v2 — Vals: 56.6%
Contributes to the score. Developer-reported. Review status: reviewed-developer.
Configuration: Muse Wonder (max); Muse Wonder developer-reported result (meta); fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-12
Developer-reported result entered in the newsroom. Same protocol as the catalogue row: the lab table reproduces cells we hold (fable-5-1 58.9, gpt-6-astra 53.5). Value read from an image of the lab table. Meta's methodology says the result source is Vals AI and the primary metric is private-test-set accuracy.
BioMysteryBench v1 — Vals: 67.04%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Vals AI BioMysteryBench harness; provider Meta; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 67.04 ± 0.37, rank 19/27. Vals notes some benchmarks may use a different provider or parameters.
Code Migration †: 62.55%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Vals AI Code Migration harness; provider Meta; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 62.55 ± 4.35, rank 6/77. Vals notes some benchmarks may use a different provider or parameters.
Excel Modeling Benchmark — Vals: 67.08%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Vals AI EMB harness; provider Meta; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 67.08 ± 2.89, rank 18/74. Vals notes some benchmarks may use a different provider or parameters.
Harvey LAB (Vals): 17.08%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Vals AI Harvey's Legal Agent Benchmark harness; provider Meta; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 17.08 ± 3.1, rank 8/79. Vals notes some benchmarks may use a different provider or parameters.
Legal Research Bench — Vals: 50.48%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Vals AI Legal Research Bench harness; provider Meta; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 50.48 ± 3.48, rank 6/78. Vals notes some benchmarks may use a different provider or parameters.
ProofBench v1.1 — Vals: 84%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Vals AI ProofBench v1.1 harness; provider Meta; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 84.0 ± 3.69, rank 12/50. Vals notes some benchmarks may use a different provider or parameters.
Public Benefits Bench — Vals: 70.23%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Vals AI Public Benefits Bench v1.1 harness; provider Meta; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 70.23 ± 1.19, rank 6/50. Vals notes some benchmarks may use a different provider or parameters.
Vibe Code Bench v1.1 — Vals: 83.09%
Contributes to the score. Source category: independent. Review status: reviewed-independent.
Configuration: Muse Wonder (max); Vals AI Vibe Code Bench v1.1 harness; provider Meta; fallback not reported
Metric: published percent
Original source · Retrieved 2026-10-15
Vals AI model page, headline accuracy 83.09 ± 2.49, rank 18/112. Vals notes some benchmarks may use a different provider or parameters.
Cite this model record
The Latent. “Muse Wonder: score and benchmark evidence.” 2026-10-26. Release tli-2026-v1.0-2026-10-26-kimi-k3-1-pro-lab-6.
Download release data and definitions (JSON) · Data usage terms