RLTasksby ulam.ai
Models / Anthropic

Anthropic / MODEL EVIDENCE

Claude Haiku 4.5

51 benchmarks · 81 observations · 56 sources · released 2025-10-15

Scores at a glance

51 benchmarks · large dot: Claude Haiku 4.5; small dots: closest models where reported

MATH Level 5 ↗96.36%
96.3696.36

Claude Haiku 4.5 · MATH level 5 · Best score (across scorers) · % ↗

1 more reported result

86.91% · Claude Haiku 4.5 · MATH level 5 · Best score (across scorers) · % · Epoch AI · MATH level 5 ↗

BALROG ↗31.2%
31.231.2

Claude Haiku 4.5 · Balrog · Average progress · % ↗

Terminal-Bench 2.0 ↗29.8%
3.178.4

Claude Haiku 4.5 · Terminal Bench · Accuracy mean · % ↗

8 more reported results

28.3% · Claude Haiku 4.5 · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

27.5% · Claude Haiku 4.5 · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

13.9% · Claude Haiku 4.5 · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

35.51% · Claude Haiku 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

29.83% · Claude Haiku 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

28.31% · Claude Haiku 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

27.53% · Claude Haiku 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

13.93% · Claude Haiku 4.5 (unknown thinking) · Terminal Bench · Accuracy mean · % · Epoch AI · Terminal Bench ↗

CritPt ↗0%
032.29

Claude Haiku 4.5 (unknown thinking) · CritPt · Accuracy · % ↗

ExploitBench ↗13.7%
13.373.8

Claude Haiku 4.5 (unknown thinking) · ExploitBench · Mean capability · % ↗

1 more reported result

13.4% · Claude Haiku 4.5 (unknown thinking) · ExploitBench · Mean capability · % · Epoch AI · ExploitBench ↗

LMCA ↗30.94%
2.7868.26

Claude Haiku 4.5 (unknown thinking) · LMCA · Score · % ↗

DTBench ↗73.6%
41.6598.93

Claude Haiku 4.5 (unknown thinking) · DTBench · Accuracy · % ↗

SciCode ↗42.25%
1.2766.9

Claude Haiku 4.5 (unknown thinking) · SciCode · Score · % ↗

LegalBench ↗81.24 %
32.9788.56

Claude Haiku 4.5 (Thinking) · LegalBench · Accuracy ↗

CaseLaw v2 ↗56.48 %
43.8479.31

Claude Haiku 4.5 (Thinking) · CaseLaw · Accuracy ↗

CorpFin v2 ↗60.61 %
60.6160.61

Claude Haiku 4.5 (Thinking) · CorpFin · Accuracy ↗

1 more reported result

60.3 % · Claude Haiku 4.5 (Nonthinking) · CorpFin · Accuracy · Vals AI · CorpFin ↗

MedCode ↗32.68 %
19.7263.57

Claude Haiku 4.5 (Thinking) · MedCode · Accuracy ↗

MedScribe ↗85.23 %
4.2791.43

Claude Haiku 4.5 (Thinking) · MedScribe · Accuracy ↗

MMLU-Pro ↗78.72 %
30.2892.38

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

MMMU-Pro ↗46.07 %
22.7790.64

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

Vibe Code Bench ↗11.39 %
092.39

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

EconBench ↗12.4242 error points
12.4212.42

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

NYU CTF Bench ↗23.5%
11.179

Claude Haiku 4.5 · NYU CTF Bench · Cotool 81-challenge subset · accuracy ↗

Closest models by score

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

18.2 percentile-point gap

17 of 26 shared benchmark comparisons

20.8 percentile-point gap

18 of 26 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. Blueprint-Bench 2 | Andon Labs ↗

    Andon Labs

  2. Arena leaderboard — WebDev evaluation ↗

    Arena

  3. SciCode Benchmark | Artificial Analysis ↗

    Artificial Analysis

  4. 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

  5. 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

  6. DTBench — Conceptual Reasoning Index ↗

    Conceptual Reasoning Index

  7. LMCA — Conceptual Reasoning Index ↗

    Conceptual Reasoning Index

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

    EconBench project · 2026

  9. ARC-AGI-1 ↗

    Epoch AI

  10. ARC-AGI-2 ↗

    Epoch AI

  11. Chess Puzzles ↗

    Epoch AI

  12. CritPt v1 ↗

    Epoch AI

  13. ExploitBench v0.1 ↗

    Epoch AI

  14. FrontierMath Tier 4 (v1, old) ↗

    Epoch AI

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

    Epoch AI

  16. GPQA Diamond ↗

    Epoch AI

  17. MATH Level 5 ↗

    Epoch AI

  18. OTIS Mock AIME 2024-2025 ↗

    Epoch AI

  19. SimpleQA Verified ↗

    Epoch AI

  20. Terminal-Bench 2.0 ↗

    Epoch AI

  21. Vending Bench 2 — methodology and published results ↗

    Epoch AI · 2026

  22. Text Arena (Coding) ↗

    Epoch AI

  23. AI benchmarks & capabilities ↗

    Epoch AI

  24. FutureSearch Evals ↗

    FutureSearch

  25. WeirdML ↗

    Håvard Tveit Ihle

  26. Scale Labs — evaluation methodology and results ↗

    Scale Labs / SEAL · 2026

  27. SWE-Bench Pro V2 — evaluation methodology and results ↗

    Scale Labs / SEAL · 2026

  28. NYU CTF Bench — Cotool evaluation, February 2026 ↗

    Cotool · 2026

  29. Terminal-Bench 2.0 leaderboard ↗

    Terminal-Bench Team

  30. AIME — evaluation methodology and results ↗

    Vals AI · 2026

  31. CaseLaw — evaluation methodology and results ↗

    Vals AI · 2026

  32. Code Migration v1 — evaluation methodology and results ↗

    Vals AI · 2026

  33. CorpFin — evaluation methodology and results ↗

    Vals AI · 2026

  34. Excel Modeling Benchmark v1 — evaluation methodology and results ↗

    Vals AI · 2026

  35. Finance Agent v2 — evaluation methodology and results ↗

    Vals AI · 2026

  36. GPQA Diamond — evaluation methodology and results ↗

    Vals AI · 2026

  37. Harvey's Legal Agent Benchmark v1 — evaluation methodology and results ↗

    Vals AI · 2026

  38. LiveCodeBench — evaluation methodology and results ↗

    Vals AI · 2026

  39. LegalBench — evaluation methodology and results ↗

    Vals AI · 2026

  40. Legal Research Bench — evaluation methodology and results ↗

    Vals AI · 2026

  41. MedCode — evaluation methodology and results ↗

    Vals AI · 2026

  42. MedQA — evaluation methodology and results ↗

    Vals AI · 2026

  43. MedScribe — evaluation methodology and results ↗

    Vals AI · 2026

  44. MGSM — evaluation methodology and results ↗

    Vals AI · 2026

  45. MMLU Pro — evaluation methodology and results ↗

    Vals AI · 2026

  46. MMMU Pro — evaluation methodology and results ↗

    Vals AI · 2026

  47. MortgageTax — evaluation methodology and results ↗

    Vals AI · 2026

  48. ProgramBench — evaluation methodology and results ↗

    Vals AI · 2026

  49. Public Benefits Bench — evaluation methodology and results ↗

    Vals AI · 2026

  50. Public Benefits Bench v1 — evaluation methodology and results ↗

    Vals AI · 2026

  51. SAGE — evaluation methodology and results ↗

    Vals AI · 2026

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

    Vals AI · 2026

  53. Tax Agent Bench — evaluation methodology and results ↗

    Vals AI · 2026

  54. TaxEval — evaluation methodology and results ↗

    Vals AI · 2026

  55. Vals Multimodal Index — evaluation methodology and results ↗

    Vals AI · 2026

  56. Vibe Code Bench — evaluation methodology and results ↗

    Vals AI · 2026

Names used in source tables

Claude Haiku 4.5 · Claude Haiku 4.5 (59k thinking) · Claude Haiku 4.5 (Nonthinking) · Claude Haiku 4.5 (Thinking) · Claude Haiku 4.5 (Thinking) (Mini-SWE-agent) · Claude Haiku 4.5 (Thinking) (OpenHands) · Claude Haiku 4.5 (low) · Claude Haiku 4.5 (unknown thinking) · Haiku 4.5 (Claude Code) xhigh · claude-haiku-4-5