Skip to content

Models / Claude Haiku 4.5

Anthropic

Claude Haiku 4.5

54 published results from 7 sources. Each card shows where the number comes from and what it does not measure. Results are never combined into one score.

Provider: Anthropic
Sources: 7
Our benchmarks: 3
Price: Not yet published

Measured by Spring Prompt

Measured by Spring Prompt · BulletBench

491

Ladder Elo · rank 2 of 16

Unit: ladder Elo, higher is better
Range: 299 to 659
Sample: 8 games
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 29 Sep 2026
Not shown: Not a business skill: chess against an engine on a real clock, used here as a speed-under-pressure signal.

Measured by Spring Prompt · BulletBench

1.3 s

Median move time · rank 3 of 16

Unit: milliseconds, lower is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 29 Sep 2026
Not shown: Not general response speed: replies are one short move.

Measured by Spring Prompt · CatalogBench

94.4%

Content quality · rank 15 of 18

Unit: % of checks, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

0.0%

Failed outputs · rank 1 of 18

Unit: % of products, lower is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not a provider outage: provider errors are retried before a product counts as failed.

Measured by Spring Prompt · CatalogBench

0.0%

Publish-ready listings · rank 15 of 18

Unit: % of products, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

0.0%

Reliably publish-ready · rank 10 of 18

Unit: % of products, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

14.3

Unsupported claims · rank 18 of 18

Unit: claims per product, lower is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

76.8%

Channel compliance · rank 18 of 18

Unit: % of products, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

100.0%

Conflicts caught · rank 1 of 18

Unit: % of conflicts, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

94.6%

Content quality · rank 15 of 18

Unit: % of checks, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

$0.0061

Cost per product · rank 4 of 18

Unit: US dollars, lower is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your cost: prices are those charged through OpenRouter on the run date.

Measured by Spring Prompt · CatalogBench

86.0%

Decision accuracy · rank 17 of 18

Unit: % of decisions, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

0.0%

Failed outputs · rank 1 of 18

Unit: % of products, lower is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not a provider outage: provider errors are retried before a product counts as failed.

Measured by Spring Prompt · CatalogBench

93.7%

Field accuracy · rank 7 of 18

Unit: % of missing fields, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

6.2%

Invented values · rank 18 of 18

Unit: % of filled values, lower is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

0.6%

Publish-ready listings · rank 18 of 18

Unit: % of products, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

0.0%

Reliably publish-ready · rank 17 of 18

Unit: % of products, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · CatalogBench

7.92

Unsupported claims · rank 18 of 18

Unit: claims per product, lower is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 30 Sep 2026
Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

Measured by Spring Prompt · ROASBench

$0.0298

Cost of a run · rank 2 of 17

Unit: US dollars, lower is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 29 Sep 2026
Not shown: Not your cost: prices are those charged through OpenRouter on the run date.

Measured by Spring Prompt · ROASBench

2

Months over budget · rank 15 of 17

Unit: months of 12, lower is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 29 Sep 2026
Not shown: Not real ad performance: the market is a deterministic simulation of one invented skincare brand.

Measured by Spring Prompt · ROASBench

17.5

Overall score · rank 16 of 17

Unit: score out of 100, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 29 Sep 2026
Not shown: Not real ad performance: the market is a deterministic simulation of one invented skincare brand.

Measured by Spring Prompt · ROASBench

−£0.21

Return on ad spend · rank 17 of 17

Unit: profit per £1 spent, higher is better
Configuration: Claude Haiku 4.5, provider default reasoning
Measured: 29 Sep 2026
Not shown: Not real ad performance: the market is a deterministic simulation of one invented skincare brand.

Reported by others

Reported by Arena (formerly LMArena)

1,419

Overall · rank 38 of 44

Unit: Arena rating, higher is better
Range: 1,413 to 1,425
Sample: 34677 votes
Configuration: Claude Haiku 4.5
Measured: 13 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,418

Overall · rank 117 of 177

Unit: Arena rating, higher is better
Range: 1,416 to 1,420
Sample: 133816 votes
Configuration: Claude Haiku 4.5
Measured: 25 Sep 2026
Not shown: Not an error rate: claims that cannot be checked on the web are skipped, and preference still carries most of the weight.

Reported by Arena (formerly LMArena)

1,420

Business, management and finance · rank 106 of 402

Unit: Arena rating, higher is better
Range: 1,415 to 1,424
Sample: 27420 votes
Configuration: Claude Haiku 4.5
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,387

Creative writing · rank 99 of 407

Unit: Arena rating, higher is better
Range: 1,382 to 1,392
Sample: 23265 votes
Configuration: Claude Haiku 4.5
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,452

Expert prompts · rank 84 of 359

Unit: Arena rating, higher is better
Range: 1,446 to 1,458
Sample: 12831 votes
Configuration: Claude Haiku 4.5
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,415

Instruction following · rank 99 of 409

Unit: Arena rating, higher is better
Range: 1,411 to 1,419
Sample: 44952 votes
Configuration: Claude Haiku 4.5
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,414

Overall · rank 122 of 409

Unit: Arena rating, higher is better
Range: 1,411 to 1,416
Sample: 140791 votes
Configuration: Claude Haiku 4.5
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Arena (formerly LMArena)

1,399

Writing, literature and language · rank 96 of 408

Unit: Arena rating, higher is better
Range: 1,395 to 1,404
Sample: 33818 votes
Configuration: Claude Haiku 4.5
Measured: 25 Sep 2026
Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

85.1%

Irrelevance detection · rank 32 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not general refusal or safety behaviour.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

95.3%

Irrelevance detection · rank 4 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not general refusal or safety behaviour.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

54.4%

Memory · rank 7 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not long-term personal memory in a product: sessions are BFCL's scripted ones.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

2.6%

Memory · rank 99 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not long-term personal memory in a product: sessions are BFCL's scripted ones.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

53.6%

Multi-turn tasks · rank 14 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not open-ended agent work: the tools and tasks are BFCL's simulated APIs.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

1.8%

Multi-turn tasks · rank 96 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not open-ended agent work: the tools and tasks are BFCL's simulated APIs.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

68.7%

Overall accuracy · rank 6 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not a neutral average: the weighting is BFCL's. Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

25.3%

Overall accuracy · rank 87 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not a neutral average: the weighting is BFCL's. Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

62.5%

Relevance detection · rank 91 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not whether the call itself was right; only that one was attempted.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

31.2%

Relevance detection · rank 106 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not whether the call itself was right; only that one was attempted.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

86.5%

Single-turn calls (curated) · rank 34 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

55.4%

Single-turn calls (curated) · rank 98 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

78.7%

Single-turn calls (user-contributed) · rank 19 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

52.5%

Single-turn calls (user-contributed) · rank 98 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not reliability on your own tools: BFCL's functions and queries are a fixed test set, and a correct call is judged by its form, not by what it achieved.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

83.5%

Web search · rank 2 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not general research quality: questions have short, checkable answers.

Reported by Berkeley Function Calling Leaderboard (BFCL) V4

19.5%

Web search · rank 45 of 109

Unit: % correct, higher is better
Configuration: claude-haiku-4.5
Measured: 16 Dec 2025
Not shown: Not general research quality: questions have short, checkable answers.

Reported by UGI Leaderboard

6.0%

Requested-length error · rank 41 of 370

Unit: % off the requested word count, lower is better
Configuration: claude-haiku-4.5
Measured: 15 Oct 2025
Not shown: Not other format limits such as character counts or bullet counts.

Reported by UGI Leaderboard

24.0%

Requested-length error · rank 217 of 370

Unit: % off the requested word count, lower is better
Configuration: claude-haiku-4.5
Measured: 15 Oct 2025
Not shown: Not other format limits such as character counts or bullet counts.

Reported by UGI Leaderboard

0.35

Style adherence · rank 178 of 370

Unit: score from 0 to 1, higher is better
Configuration: claude-haiku-4.5
Measured: 15 Oct 2025
Not shown: Not brand voice on your own examples: UGI's prompts are private and lean towards creative writing.

Reported by UGI Leaderboard

0.39

Style adherence · rank 43 of 370

Unit: score from 0 to 1, higher is better
Configuration: claude-haiku-4.5
Measured: 15 Oct 2025
Not shown: Not brand voice on your own examples: UGI's prompts are private and lean towards creative writing.

Reported by UGI Leaderboard

52.2

Writing score · rank 145 of 370

Unit: score out of 100, higher is better
Configuration: claude-haiku-4.5
Measured: 15 Oct 2025
Not shown: Not business copy quality: UGI's prompts are private and lean towards creative writing, and models that often refuse get no score.

Reported by UGI Leaderboard

49.1

Writing score · rank 162 of 370

Unit: score out of 100, higher is better
Configuration: claude-haiku-4.5
Measured: 15 Oct 2025
Not shown: Not business copy quality: UGI's prompts are private and lean towards creative writing, and models that often refuse get no score.

Reported by Vectara Hallucination Leaderboard

99.5%

Answer rate · rank 53 of 108

Unit: % of documents, higher is better
Configuration: Claude Haiku 4.5
Measured: 22 Sep 2026
Not shown: Not a quality score: a low rate usually means content filters were triggered, and hallucination rates are measured on answered documents only.

Reported by Vectara Hallucination Leaderboard

9.8%

Hallucination rate · rank 58 of 108

Unit: % of summaries, lower is better
Configuration: Claude Haiku 4.5
Measured: 22 Sep 2026
Not shown: Not errors in open questions or other tasks: only summarisation, judged by Vectara's own model (HHEM-2.3), not by people, on news-style documents rather than your data.

Compare Claude Haiku 4.5 with