Anthropic·released 2025-10-151 source
Claude Haiku 4.5 benchmark scores: 16 benchmarks tracked. 31% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 0 of 16 benchmarks·best result 96.36 on MATH level 5 (reproduced)·5 of 16 independently reproduced·$1/$5 per M tokens
Consensus: LiteLLM · Cross-check: models.dev, OpenRouter · See every model’s pricing →
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| MATH level 5 | 96.36accuracy (%) | reproduced· optimizedT1 | 2025-10-15Epoch AI |
| OTIS Mock AIME 2024-2025 | 66.63accuracy (%) | reproduced· optimizedT1 | 2025-10-15Epoch AI |
| GPQA diamond | 61.62accuracy (%) | reproduced· optimizedT1 | 2025-10-15Epoch AI |
| DTBench | 56accuracy (%) | unverifiedT1 | 2025-10-15https://conceptualreasoning.ai/dtbench |
| ARC-AGI | 47.67accuracy (%) | unverified· optimizedT2 | 2025-10-15 |
| DeepResearch Bench | 45.5accuracy (%) | unverifiedT1 | 2025-10-15https://evals.futuresearch.ai/#drb |
| WeirdML | 45.4accuracy (%) | unverifiedT1 | 2025-10-15https://htihle.github.io/weirdml.html |
| LMCA | 36.4accuracy (%) | unverifiedT1 | 2025-10-15https://conceptualreasoning.ai/lmca |
| Terminal Bench | 35.51accuracy (%) | unverifiedT1 | 2025-10-15https://www.tbench.ai/leaderboard/terminal-bench/2.0 |
| Balrog | 31.2accuracy (%) | unverifiedT1 | 2025-10-15Balrog Leaderboard |
| ExploitBench | 13.7accuracy (%) | unverifiedT2 | 2025-10-15 |
| SimpleQA Verified | 13.2accuracy (%) | unverifiedT2 | 2025-10-15 |
| FrontierMath-2025-02-28-Private | 10.36accuracy (%) | reproduced· optimizedT1 | 2025-10-15Epoch AI |
| ARC-AGI-2 | 4.03accuracy (%) | unverifiedT2 | 2025-10-15 |
| FrontierMath-Tier-4-2025-07-01-Private | 3.47accuracy (%) | reproducedT1 | 2025-10-15Epoch AI |
| Chess Puzzles | 3.2accuracy (%) | unverifiedT2 | 2025-10-15 |
Claims drawn from cited facts, not live model generation.
This model was developed by Anthropic and released in 2025.
2 cited facts
This model has scores on 16 tracked benchmarks and currently leads 0 of them. The record’s evidence status is independently reproduced.
3 cited facts
Its mean gap to the best recorded score across its tracked benchmarks is 43.29 points.
1 cited fact
6 facts cross-checked across data sources: 3 corroborated, 2 single-source, 1 disputed.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks, litellm_prices, modelsdev_models, openrouter_models
Claude Haiku 4.5 is an AI model developed by Anthropic, released 2025-10-15. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
Claude Haiku 4.5 has recorded scores on 16 benchmarks, each shown with its evidence status.
Claude Haiku 4.5 has recorded scores on 16 benchmarks — GPQA diamond, MATH level 5, OTIS Mock AIME 2024-2025, DTBench, WeirdML, LMCA, and 10 more. The full table above shows each score with its evidence status.
5 of 16 recorded scores (31%) are independently reproduced rather than self-reported by the lab.
Claude Haiku 4.5 has 16 tracked claims: 5 independently reproduced, 11 unverified.
Listed API pricing: $1 per million input tokens, $5 per million output tokens. See the pricing block for the full breakdown.