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.

← Models

Claude Opus 4.7

Anthropic·released 2026-04-16

independently reproduced

Scored on 21 benchmarks, leading 3. 33% of its scores are independently reproduced — the rest are self-reported or unverified.

21
Benchmarks scored
3
SOTA leads
97.8
Best score
33%
Reproduced
7 of 21

Leads on: GSO-Bench, SWE-Bench verified, Terminal Bench

Benchmark fingerprint

Every score, ranked, colored by evidence status.

  • OTIS Mock AIME 2024-2025
    97.8
  • ARC-AGI
    93.5
  • Terminal Bench
    90.2
  • GPQA diamond
    86.87
  • SWE-Bench verified
    83.47
  • WeirdML
    76.4
  • ARC-AGI-2
    75.8
  • FrontierMath-Tiers-1-3-v2-Private
    70.18
  • CursorBench
    64.8
  • SimpleBench
    55.48
  • SimpleQA Verified
    50.6
  • GSO-Bench
    44.12
  • GBAEval
    43.81
  • APEX-Agents
    33.9
  • HLE
    32.98
  • FrontierMath-Tier-4-v2-Private
    31.71
  • Chess Puzzles
    30
  • PostTrainBench
    28.56
  • ExploitBench
    28
  • OSWorld 2.0
    18.2
  • CritPt
    12

All benchmark scores

BenchmarkScoreEvidence statusMeasured
OTIS Mock AIME 2024-202597.8accuracy (%)reproduced· optimized2026-04-16
ARC-AGI93.5accuracy (%)unverified· optimized2026-04-16
Terminal Bench90.2accuracy (%)unverified· optimized2026-04-16
GPQA diamond86.87accuracy (%)reproduced· optimized2026-04-16
SWE-Bench verified83.47accuracy (%)reproduced· optimized2026-04-16
WeirdML76.4accuracy (%)unverified2026-04-16
ARC-AGI-275.8accuracy (%)unverified2026-04-16
FrontierMath-Tiers-1-3-v2-Private70.18accuracy (%)reproduced2026-04-16
CursorBench64.8accuracy (%)unverified2026-04-16
SimpleBench55.48accuracy (%)unverified2026-04-16
SimpleQA Verified50.6accuracy (%)reproduced2026-04-16
GSO-Bench44.12accuracy (%)unverified2026-04-16
GBAEval43.81accuracy (%)unverified2026-04-16
APEX-Agents33.9accuracy (%)unverified2026-04-16
HLE32.98accuracy (%)unverified2026-04-16
FrontierMath-Tier-4-v2-Private31.71accuracy (%)reproduced2026-04-16
Chess Puzzles30accuracy (%)reproduced2026-04-16
PostTrainBench28.56accuracy (%)unverified2026-04-16
ExploitBench28accuracy (%)unverified2026-04-16
OSWorld 2.018.2accuracy (%)unverified2026-04-16
CritPt12accuracy (%)unverified2026-04-16
21 source records · epoch_benchmarks. Every score is evidence-linked.

Follow the record

More from Anthropic

  • Claude Fable 599.72
  • Claude Opus 4.898.33
  • Claude Sonnet 4.597.73
  • Claude Haiku 4.596.36
  • Claude Opus 4.694.44
  • Claude 3.7 Sonnet91.16

Benchmarks it leads

  • GSO-Bench
  • SWE-Bench verified
  • Terminal Bench