RLTasksby ulam.ai
Models / Anthropic

Anthropic / MODEL EVIDENCE

Claude Opus 4.5

59 benchmarks · 121 observations · 66 sources · released 2025-11-24

Scores at a glance

59 benchmarks · large dot: Claude Opus 4.5; small dots: closest models where reported

GPQA Diamond ↗86.05%
86.0586.05

Claude Opus 4.5 (32k thinking) · GPQA diamond · Best score (across scorers) · % ↗

4 more reported results

85.48% · Claude Opus 4.5 (16k thinking) · GPQA diamond · Best score (across scorers) · % · Epoch AI · GPQA diamond ↗

80.68% · Claude Opus 4.5 (no thinking) · GPQA diamond · Best score (across scorers) · % · Epoch AI · GPQA diamond ↗

85.86 % · Claude Opus 4.5 (Thinking) · GPQA Diamond · Accuracy · Vals AI · GPQA Diamond ↗

79.55 % · Claude Opus 4.5 (Nonthinking) · GPQA Diamond · Accuracy · Vals AI · GPQA Diamond ↗

FrontierMath Tiers 1-3 (v1, old) ↗20.69%
20.6920.69

Claude Opus 4.5 (no thinking) · FrontierMath-2025-02-28-Private · Best score (across scorers) · % ↗

2 more reported results

20.69% · Claude Opus 4.5 (32k thinking) · FrontierMath-2025-02-28-Private · Best score (across scorers) · % · Epoch AI · FrontierMath-2025-02-28-Private ↗

20.34% · Claude Opus 4.5 (16k thinking) · FrontierMath-2025-02-28-Private · Best score (across scorers) · % · Epoch AI · FrontierMath-2025-02-28-Private ↗

FrontierMath Tier 4 (v1, old) ↗4.17%
4.174.17

Claude Opus 4.5 (no thinking) · FrontierMath-Tier-4-2025-07-01-Private · Best score (across scorers) · % ↗

2 more reported results

4.17% · Claude Opus 4.5 (32k thinking) · FrontierMath-Tier-4-2025-07-01-Private · Best score (across scorers) · % · Epoch AI · FrontierMath-Tier-4-2025-07-01-Private ↗

2.08% · Claude Opus 4.5 (16k thinking) · FrontierMath-Tier-4-2025-07-01-Private · Best score (across scorers) · % · Epoch AI · FrontierMath-Tier-4-2025-07-01-Private ↗

SimpleQA Verified ↗45.7%
676

Claude Opus 4.5 (32k thinking) · SimpleQA Verified · Best score (across scorers) · % ↗

ARC-AGI-1 ↗80%
8080

Claude Opus 4.5 (64k thinking) · ARC-AGI · Score · % ↗

5 more reported results

75.83% · Claude Opus 4.5 (32k thinking) · ARC-AGI · Score · % · Epoch AI · ARC-AGI ↗

75.8% · Claude Opus 4.5 (32k thinking) · ARC-AGI · Score · % · Epoch AI · ARC-AGI ↗

72% · Claude Opus 4.5 (16k thinking) · ARC-AGI · Score · % · Epoch AI · ARC-AGI ↗

58.67% · Claude Opus 4.5 (8k thinking) · ARC-AGI · Score · % · Epoch AI · ARC-AGI ↗

40% · Claude Opus 4.5 (no thinking) · ARC-AGI · Score · % · Epoch AI · ARC-AGI ↗

BALROG ↗43.5%
43.543.5

Claude Opus 4.5 (no thinking) · Balrog · Average progress · % ↗

2 more reported results

43% · Claude Opus 4.5 (64k thinking) · Balrog · Average progress · % · Epoch AI · Balrog ↗

43% · Claude Opus 4.5 (unknown thinking) · Balrog · Average progress · % · Epoch AI · Balrog ↗

Cybench ↗82%
6082

Claude Opus 4.5 (unknown thinking) · Cybench · Unguided % Solved · % ↗

GeoBench ↗75%
7588

Claude Opus 4.5 (no thinking) · GeoBench · ACW Country % · % ↗

GSO ↗26.47%
1.344.12

Claude Opus 4.5 (unknown thinking) · GSO-Bench · Score OPT@1 · % ↗

1 more reported result

26.5% · Claude Opus 4.5 (no thinking) · GSO-Bench · Score OPT@1 · % · Epoch AI · GSO-Bench ↗

Terminal-Bench 2.0 ↗63.15%
3.0784.72

Claude Opus 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % ↗

9 more reported results

59.1% · Claude Opus 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

58.43% · Claude Opus 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

57.75% · Claude Opus 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

54.33% · Claude Opus 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

52.13% · Claude Opus 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

51.91% · Claude Opus 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

51.69% · Claude Opus 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

63.1% · Claude Opus 4.5 (no thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

59.1% · Claude Opus 4.5 (128k thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

VPCT ↗40%
4040

Claude Opus 4.5 (32k thinking) · VPCT · Correct · % ↗

GDPval ↗45.5%
9.949.7

Claude Opus 4.5 (no thinking) · GDPval · Win Rate (%) · % ↗

CL-bench ↗21.1%
11.427.9

Claude Opus 4.5 (unknown thinking) · CL-bench · Overall · % ↗

METR task performance & time horizons ↗74.97%
10.185.2

Claude Opus 4.5 (16k thinking) · METR task performance & time horizons · Average task accuracy ↗

3 more reported results

73.01% · Claude Opus 4.5 (no thinking) · METR task performance & time horizons · Average task accuracy · Epoch AI · METR Time Horizons ↗

292.99459 minutes · Claude Opus 4.5 (no thinking) · METR task performance & time horizons · 50% success task horizon · METR-Horizon-v1.1 · Epoch AI · METR Autonomy / Task-Horizon Suite · public result data ↗

288.90495 minutes · Claude Opus 4.5 (16k thinking) · METR task performance & time horizons · 50% success task horizon · Epoch AI · METR Autonomy / Task-Horizon Suite · public result data ↗

ProofBench ↗36%
0100

Claude Opus 4.5 (unknown thinking) · ProofBench · Accuracy · % ↗

LMCA ↗44.46%
2.7868.26

Claude Opus 4.5 (high) · LMCA · Score · % ↗

DTBench ↗89.87%
41.6598.93

Claude Opus 4.5 (high) · DTBench · Accuracy · % ↗

EnigmaEval ↗11.91%
0.3839.28

Claude Opus 4.5 (unknown thinking) · EnigmaEval · Accuracy · % ↗

Vals RSI Index ↗0.22 reference score
00.46

Claude Opus 4.5 (Nonthinking) (high) · Compression · reference-anchored score ↗

3 more reported results

0.18 reference score · Claude Opus 4.5 (Nonthinking) (high) · LM training · reference-anchored score · Vals RSI Index · LM training ↗

0.19 reference score · Claude Opus 4.5 (Nonthinking) (high) · Harness engineering · reference-anchored score · Vals RSI Index · Harness engineering ↗

0.00 reference score · Claude Opus 4.5 (Nonthinking) (high) · Post-training · reference-anchored score · Vals RSI Index · Post-training ↗

MedCode ↗49.16 %
49.1649.16

Claude Opus 4.5 (Thinking) · MedCode · Accuracy ↗

1 more reported result

45.17 % · Claude Opus 4.5 (Nonthinking) · MedCode · Accuracy · Vals AI · MedCode ↗

MedScribe ↗85.32 %
85.3285.32

Claude Opus 4.5 (Thinking) · MedScribe · Accuracy ↗

1 more reported result

83.25 % · Claude Opus 4.5 (Nonthinking) · MedScribe · Accuracy · Vals AI · MedScribe ↗

MMLU-Pro ↗87.26 %
87.2687.26

Claude Opus 4.5 (Thinking) · MMLU Pro · Accuracy ↗

1 more reported result

85.59 % · Claude Opus 4.5 (Nonthinking) · MMLU Pro · Accuracy · Vals AI · MMLU Pro ↗

MMMU-Pro ↗82.95 %
82.9582.95

Claude Opus 4.5 (Thinking) · MMMU Pro · Accuracy ↗

1 more reported result

81.1 % · Claude Opus 4.5 (Nonthinking) · MMMU Pro · Accuracy · Vals AI · MMMU Pro ↗

Vibe Code Bench ↗20.63 %
092.39

Claude Opus 4.5 (Thinking) (OpenHands) · Vibe Code Bench · original task set ↗

EconBench ↗16.7214 error points
16.7216.72

claude-opus-4-5 · EconBench · EconBench prototype · risk-consistency error · published prototype observations ↗

UnsolvedMath ↗5 / 5
55

Claude Opus 4.5 · UnsolvedMath · Literature search · Ulam · January 31, 2026 research assessment ↗

3 more reported results

2 / 5 · Claude Opus 4.5 · UnsolvedMath · Theorem proving · Ulam · January 31, 2026 research assessment · Open Mathematical Problems as an AI Reasoning Benchmark ↗

1 / 5 · Claude Opus 4.5 · UnsolvedMath · Proof checking · Ulam · January 31, 2026 research assessment · Open Mathematical Problems as an AI Reasoning Benchmark ↗

0 / 1 · Claude Opus 4.5 · UnsolvedMath · Formalization · Ulam · January 31, 2026 research assessment · Open Mathematical Problems as an AI Reasoning Benchmark ↗

AlgoTune ↗1.77 ×
1.312.05

Claude Opus 4.5 (high) · Harmonic-mean speedup ↗

Closest models by score

Similar positions on shared benchmark leaderboards. Smaller gaps are closer.

17.3 percentile-point gap

10 of 12 shared benchmark comparisons

18.2 percentile-point gap

8 of 12 shared benchmark comparisons

How similarity is measured

Mean absolute gap in percentile positions on shared benchmark settings; lower is closer. At least 60% overlap is required, with a minimum of three when available. Independent evaluations are used when there are at least three; otherwise published report comparisons are used. Repeated model observations within one setting are averaged before ranking. Writing-style traits and composite indices are excluded. This describes similarity, not an overall quality ranking.

Filter comparisons to this model →

Sources

Every plotted observation links to its evaluation. Reported configurations, benchmark versions and task subsets can differ; similarity describes shared scores, not training lift.

  1. ARC Prize - Leaderboard ↗

    ARC Prize Foundation

  2. Arena leaderboard — WebDev evaluation ↗

    Arena

  3. OfficeQA Pro: An Enterprise Benchmark for End-to-End Grounded Reasoning — published result table ↗

    Opsahl-Ong, Krista, Singhvi, Arnav, Collins, Jasmine, Zhou, Ivan, Wang, Cindy, Baheti, Ashutosh et al. · 2026

    All 13 authors

    Opsahl-Ong, Krista, Singhvi, Arnav, Collins, Jasmine, Zhou, Ivan, Wang, Cindy, Baheti, Ashutosh, Oertell, Owen, Portes, Jacob, Havens, Sam, Elsen, Erich, Bendersky, Michael, Zaharia, Matei, Chen, Xing

  4. TasteVal: Measuring the Experimental Research Taste of AI Systems Against Human Experts — published result table ↗

    Jaffe, Oliver, Sherburn, Dane · 2026

  5. System Card: Claude Opus 4.5 ↗

    Anthropic · 2025

  6. BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games ↗

    Paglieri, Davide, Cupiał, Bartłomiej, Coward, Samuel, Piterbarg, Ulyana, Wolczyk, Maciej, Khan, Akbir et al. · 2024

    All 13 authors

    Paglieri, Davide, Cupiał, Bartłomiej, Coward, Samuel, Piterbarg, Ulyana, Wolczyk, Maciej, Khan, Akbir, Pignatelli, Eduardo, Kuciński, Łukasz, Pinto, Lerrel, Fergus, Rob, Foerster, Jakob Nicolaus, Parker-Holder, Jack, Rocktäschel, Tim

  7. DTBench — Conceptual Reasoning Index ↗

    Conceptual Reasoning Index

  8. LMCA — Conceptual Reasoning Index ↗

    Conceptual Reasoning Index

  9. EconBench: Economic choice by language models — claude-opus-4-5 prototype metrics ↗

    EconBench project · 2026

  10. Algotune — methodology and published results ↗

    Epoch AI · 2026

  11. ARC-AGI-1 ↗

    Epoch AI

  12. ARC-AGI-2 ↗

    Epoch AI

  13. Chess Puzzles ↗

    Epoch AI

  14. CL-bench ↗

    Epoch AI

  15. Earthborne Rangers (EBR-bench) ↗

    Epoch AI

  16. FrontierMath Tier 4 (v1, old) ↗

    Epoch AI

  17. FrontierMath Tier 4 (v2) ↗

    Epoch AI

  18. FrontierMath Tiers 1-3 (v1, old) ↗

    Epoch AI

  19. FrontierMath Tiers 1-3 (v2) ↗

    Epoch AI

  20. Furniture Assembly Benchmark ↗

    Epoch AI

  21. GDPval ↗

    Epoch AI

  22. GeoBench ↗

    Epoch AI

  23. GPQA Diamond ↗

    Epoch AI

  24. Humanity's Last Exam ↗

    Epoch AI

  25. METR Time Horizons ↗

    Epoch AI

  26. Mystery Game Puzzles ↗

    Epoch AI

  27. OTIS Mock AIME 2024-2025 ↗

    Epoch AI

  28. Remote Labor Index ↗

    Epoch AI

  29. SimpleQA Verified ↗

    Epoch AI

  30. SimpleBench ↗

    Epoch AI

  31. SWE-bench Verified ↗

    Epoch AI

  32. Terminal-Bench 2.0 ↗

    Epoch AI

  33. Capabilities & benchmarking — retained external evaluation records ↗

    Epoch AI · 2026

  34. Vending Bench 2 — methodology and published results ↗

    Epoch AI · 2026

  35. AI benchmarks & capabilities ↗

    Epoch AI

  36. FutureSearch Evals ↗

    FutureSearch

  37. GSO: Challenging Software Optimization Tasks for Evaluating SWE-Agents ↗

    Shetty, Manish, Jain, Naman, Liu, Jinjian, Kethanaboyina, Vijay, Sen, Koushik, Stoica, Ion · 2025

  38. WeirdML ↗

    Håvard Tveit Ihle

  39. EnigmaEval ↗

    Scale Labs / SEAL

  40. Scale Labs — evaluation methodology and results ↗

    Scale Labs / SEAL · 2026

  41. Measuring AI Ability to Complete Long Software Tasks ↗

    METR

  42. Metr Autonomy Task Horizon Suite — methodology and published results ↗

    METR · 2026

  43. Introducing Claude Opus 4.5 ↗

    Anthropic

  44. Terminal-Bench 2.0 leaderboard ↗

    Terminal-Bench Team

  45. Open Mathematical Problems as an AI Reasoning Benchmark ↗

    Przemyslaw Chojecki · 2026

  46. AIME — evaluation methodology and results ↗

    Vals AI · 2026

  47. CaseLaw — evaluation methodology and results ↗

    Vals AI · 2026

  48. CorpFin — evaluation methodology and results ↗

    Vals AI · 2026

  49. GPQA Diamond — evaluation methodology and results ↗

    Vals AI · 2026

  50. LiveCodeBench — evaluation methodology and results ↗

    Vals AI · 2026

  51. LegalBench — evaluation methodology and results ↗

    Vals AI · 2026

  52. MedCode — evaluation methodology and results ↗

    Vals AI · 2026

  53. MedQA — evaluation methodology and results ↗

    Vals AI · 2026

  54. MedScribe — evaluation methodology and results ↗

    Vals AI · 2026

  55. MGSM — evaluation methodology and results ↗

    Vals AI · 2026

  56. MMLU Pro — evaluation methodology and results ↗

    Vals AI · 2026

  57. MMMU Pro — evaluation methodology and results ↗

    Vals AI · 2026

  58. MortgageTax — evaluation methodology and results ↗

    Vals AI · 2026

  59. Poker Agent — evaluation methodology and results ↗

    Vals AI · 2026

  60. ProofBench — evaluation methodology and results ↗

    Vals AI · 2026

  61. Vals RSI Index — evaluation methodology and results ↗

    Vals AI · 2026

  62. SAGE — evaluation methodology and results ↗

    Vals AI · 2026

  63. SWE-bench Verified — evaluation methodology and results ↗

    Vals AI · 2026

  64. TaxEval — evaluation methodology and results ↗

    Vals AI · 2026

  65. Vibe Code Bench — evaluation methodology and results ↗

    Vals AI · 2026

  66. VPCT benchmark results announcement ↗

    @ChaseBrowe32432

Names used in source tables

Claude Opus 4.5 · Claude Opus 4.5 (128k thinking) · Claude Opus 4.5 (16k thinking) · Claude Opus 4.5 (32k thinking) · Claude Opus 4.5 (48k thinking) · Claude Opus 4.5 (59k thinking) · Claude Opus 4.5 (64k thinking) · Claude Opus 4.5 (8k thinking) · Claude Opus 4.5 (Nonthinking) · Claude Opus 4.5 (Nonthinking) (high) · Claude Opus 4.5 (Thinking) · Claude Opus 4.5 (Thinking) (Mini-SWE-agent) · Claude Opus 4.5 (Thinking) (OpenHands) · Claude Opus 4.5 (high) · Claude Opus 4.5 (low) · Claude Opus 4.5 (no thinking) · Claude Opus 4.5 (unknown thinking) · Opus 4.5 (24k thinking budget) · claude-opus-4-5 · claude-opus-4-5 (high)