tensor.news
Integrity IndexModelsCompareLabsPapersMethodologyAbout
tensor.news

© 2026

Evidence

  • Integrity Index
  • Model Claim Ledger
  • Compare

Entities

  • Models
  • Benchmarks
  • Labs
  • Papers

Method

  • Methodology
  • About
  • Privacy

Machine

  • Feed (Atom)
  • llms.txt
  • llms-full.txt
  • Sitemap

tensor.news grades benchmarks and tags model claims by evidence status. Every fact carries a source record and provenance. Scores are framed as task performance under a disclosed harness, never as deployed capability.

← Compare

Claude Opus 4.5 vs Claude Opus 4.6

Verdict

Claude Opus 4.6 leads 18–1 across 19 shared benchmarks.

Claude Opus 4.5 1 · Claude Opus 4.6 18 · higher isn't always better — see caveats.

Per-benchmark head-to-head

Claude Opus 4.5Claude Opus 4.6
  • 18.4
    APEX-Agents
    31.7
  • 80
    ARC-AGI
    94
  • 37.64
    ARC-AGI-2
    69.17
  • 12
    Chess Puzzles
    17
  • 21.1
    CL-bench
    20.7
  • 82
    Cybench
    93
  • 4.88
    FrontierMath-Tier-4-v2-Private
    26.83
  • 34.39
    FrontierMath-Tiers-1-3-v2-Private
    65.96
  • 81.4
    GPQA diamond
    87.37
  • 26.5
    GSO-Bench
    41.2
  • 21.43
    HLE
    31.13
  • 86.1
    OTIS Mock AIME 2024-2025
    94.44
  • 17.29
    PostTrainBench
    24.82
  • 3.75
    Remote Labor Index
    4.17
  • 54.4
    SimpleBench
    61.12
  • 41.8
    SimpleQA Verified
    46.49
  • 76.65
    SWE-Bench verified
    78.72
  • 63.1
    Terminal Bench
    79.8
  • 63.7
    WeirdML
    77.95
A higher number is not always a better model. Each score is task performance under a disclosed harness, and some of these benchmarks are saturated or scored under mixed harnesses. Follow any benchmark to its integrity grade before reading a win as decisive.

Follow the record

These models

  • Claude Opus 4.586.1
  • Claude Opus 4.694.44