Shared benchmarks

20
Shared benchmarks for Claude Opus 4.5 and Claude Opus 4.6
BenchmarkClaude Opus 4.5Claude Opus 4.6
  • 87.0%
  • 91.3%
Evaluation details for GPQA Diamond

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on Graduate-Level Google-Proof Q&A Diamond

Claude Opus 4.6

Record for Claude Opus 4.6 on Graduate-Level Google-Proof Q&A Diamond
  • 80.7%
  • 88.4%
Evaluation details for GPQA Diamond — Epoch AI harness

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on Graduate-Level Google-Proof Q&A Diamond — Epoch AI harness

Claude Opus 4.6

  • 79.5%
  • 85.9%thinking
  • 89.6%
Evaluation details for GPQA Diamond — Vals AI harness

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on Graduate-Level Google-Proof Q&A Diamond — Vals AI harness
Record for Claude Opus 4.5 on Graduate-Level Google-Proof Q&A Diamond — Vals AI harness

Claude Opus 4.6

Record for Claude Opus 4.6 on Graduate-Level Google-Proof Q&A Diamond — Vals AI harness
  • 30.8%
  • 40.0%
Evaluation details for Humanity's Last Exam Full set — No tools

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on Humanity's Last Exam Full set — No tools

Claude Opus 4.6

Record for Claude Opus 4.6 on Humanity's Last Exam Full set — No tools
  • 14.2%
  • 25.2%thinking
  • 19.0%Non-Thinking
  • 34.4%thinking-max
Evaluation details for Humanity's Last Exam Full set — Scale AI evaluation

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

  • 13.9%
  • 26.3%thinking
  • 19.4%Non-Thinking
  • 36.2%thinking-max
Evaluation details for Humanity's Last Exam Text only — Scale AI evaluation

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

  • 69.8%
  • 76.8%
Evaluation details for MCP Atlas Public, April 2026 update

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Claude Opus 4.6

  • 85.6%
  • 87.3%thinking
  • 89.1%
Evaluation details for MMLU-Pro Original — Vals AI harness

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on MMLU-Pro Original — Vals AI harness

Claude Opus 4.6

  • 81.1%
  • 82.9%thinking
  • 83.9%
Evaluation details for MMMU-Pro Vals AI harness

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on MMMU-Pro Vals AI harness
Record for Claude Opus 4.5 on MMMU-Pro Vals AI harness

Claude Opus 4.6

Record for Claude Opus 4.6 on MMMU-Pro Vals AI harness
  • 66.3%
  • 72.7%
Evaluation details for OSWorld Verified

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on OSWorld Verified

Claude Opus 4.6

Record for Claude Opus 4.6 on OSWorld Verified
  • 11.2%
  • 27.2%
Evaluation details for Riemann-bench October 2026 leaderboard

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on Riemann-bench October 2026 leaderboard

Claude Opus 4.6

Record for Claude Opus 4.6 on Riemann-bench October 2026 leaderboard
  • 70.7%
  • 72.0%
Evaluation details for SWE-bench Multilingual — mini-swe-agent

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on SWE-bench Multilingual — mini-swe-agent

Claude Opus 4.6

Record for Claude Opus 4.6 on SWE-bench Multilingual — mini-swe-agent
  • 80.9%
  • 80.8%
Evaluation details for SWE-bench Verified

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on SWE-bench Verified

Claude Opus 4.6

Record for Claude Opus 4.6 on SWE-bench Verified
  • 76.7%
  • 75.6%
Evaluation details for SWE-bench Verified — Epoch AI harness

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on SWE-bench Verified — Epoch AI harness

Claude Opus 4.6

Record for Claude Opus 4.6 on SWE-bench Verified — Epoch AI harness
  • 74.4%medium
  • 76.8%high
  • 75.6%
Evaluation details for SWE-bench Verified — mini-swe-agent

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on SWE-bench Verified — mini-swe-agent
Record for Claude Opus 4.5 on SWE-bench Verified — mini-swe-agent

Claude Opus 4.6

Record for Claude Opus 4.6 on SWE-bench Verified — mini-swe-agent
  • 76.4%
  • 78.2%
Evaluation details for SWE-bench Verified — Vals AI mini-swe-agent

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Claude Opus 4.6

  • 59.8%
  • 65.4%
Evaluation details for Terminal-Bench 2.0

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on Terminal-Bench 2.0

Claude Opus 4.6

Record for Claude Opus 4.6 on Terminal-Bench 2.0
  • 58.4%
  • 53.9%thinking
  • 58.4%
Evaluation details for Terminal-Bench 2.0 — Vals AI Terminus 2

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

Claude Opus 4.5

Record for Claude Opus 4.5 on Terminal-Bench 2.0 — Vals AI Terminus 2
Record for Claude Opus 4.5 on Terminal-Bench 2.0 — Vals AI Terminus 2

Claude Opus 4.6

Record for Claude Opus 4.6 on Terminal-Bench 2.0 — Vals AI Terminus 2
  • 20.6%
  • 57.6%
  • 53.5%thinking
Evaluation details for Vibe Code Bench 1.1 — OpenHands, October 2026 evaluation

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.

  • 24.7%
  • 27.3%
Evaluation details for τ³-bench τ³ — Banking, AllTools retrieval, v1.0.1 grading

The recorded benchmark version, metric, and evaluator sets match. Evaluation methodology is not recorded by the registry.