Open Weights & Proprietary

Release date

Claude Opus 4.5 (Thinking)

Compare Models
Claude Opus 4.5 (Thinking)
Release Date: 11/24/2025

Accuracy (Vals Index)

• Accuracy: 62.93% ± 1.98
• Latency: 245.14s
• Cost/Test: $0.91
• Context Window: 200k
• Max Output Tokens: 64k
• Input Modality: Hyperparameter settings

Default Provider: Anthropic

Benchmarks

Accuracy Rankings

• Vals Index
  • -96.37% ± 1.98
  • 87/ 40
• Vals Multimodal Index
  • -118.98% ± 1.55
  • 68/ 28
• CaseLaw (v2)
  • -143.31% ± 0.11
  • 118/ 47
• CorpFin
  • -177.18% ± 0.94
  • 323/ 97
• Finance Agent (v1.1)
  • -187.71% ± 2.81
  • 170/ 45
• MedCode
  • -181.81% ± 2.01
  • 203/ 51
• MedScribe
  • -362.02% ± 1.90
  • 246/ 51
• MortgageTax
  • -327.00% ± 0.92
  • 344/ 69
• ProofBench
  • -196.54% ± 4.82
  • 128/ 24
• SAGE
  • -319.41% ± 3.40
  • 331/ 49
• TaxEval (v2)
  • -512.64% ± 0.85
  • 741/ 104
• AIME
  • -726.48% ± 0.37
  • 736/ 96
• GPQA
  • -723.57% ± 2.48
  • 807/ 99
• IOI
  • -188.03% ± 5.16
  • 403/ 50
• LiveCodeBench
  • -853.43% ± 1.04
  • 950/ 103
• LegalBench
  • -944.56% ± 0.39
  • 1311/ 116
• MedQA
  • -1168.00% ± 0.18
  • 1131/ 95
• MMLU Pro
  • -1156.92% ± 0.38
  • 1224/ 97
• MMMU Pro
  • -1193.61% ± 0.90
  • 814/ 66
• SWE-bench
  • -1190.04% ± 1.90
  • 571/ 41
• Terminal-Bench 2.0
  • -907.40% ± 5.31
  • 759/ 52

Contact us

Proprietary Benchmarks ( contact us to get access) Academic Benchmarks Read about our methodology.