Compare Models

Wonder Full

Wonder Full is a model from Meta, released October 12, 2026. It ranks #5 of 47 models on the Vals Index with 64.57%. Its best result is #4 of 79 on Harvey's Legal Agent Benchmark.

Release Date: Oct 12, 2026

Developer Meta  🇺🇸
Context Window 1M
Max Output Tokens 131k
Token Costs (in/out) $4.00/20.00
Weights Private
Input Modalities

Accuracy

64.57 % ± 1.07

Cost / Test (Vals Index)

$ 8.559

Latency

56 min 36 s

Fallback Rate 0.00%
Refusal Rate 0.04%
Vals Index
BenchmarksAccuracyRankings
Vals Index

64.57%

±1.07
5/47
Code Migration

61.03%

±4.42
7/77
CyberBench v1.1

73.63%

±5.56
12/47
EMB

71.92%

±2.66
9/74
Finance Agent (v2)

59.06%

±0.21
5/78
Legal Research Bench

53.37%

±3.47
5/78
MysteryMechanism

35.13%

±3.21
11/27
ProofBench v1.1

94.00%

±2.39
9/50
Public Benefits Bench v1.1

73.75%

±1.15
3/50
Tax Agent Bench

43.00%

±3.57
5/69
Vibe Code Bench v1.1

86.00%

±1.93
12/112
BioMysteryBench

73.33%

±0.64
8/27
Harvey's Legal Agent Benchmark

21.67%

±3.54
4/79
IOI

79.83%

±2.99
11/44
ProgramBench

0.50%

±0.50
27/64
Terminal-Bench 4.0

58.08%

±3.31
5/47
Proprietary BenchmarksAcademic BenchmarksIndustry Partners
Vals
Default Provider : Meta
Temperature: 1
Top P: Default
Top K: Default
Max Output Tokens: 131,072
Reasoning Effort: max