Models / Claude Haiku 4.5

Anthropic

Claude Haiku 4.5

Claude Haiku 4.5 trails most of the field on what it has been measured on; worth a look for simple, high-volume work where cost matters most. The cheapest model of similar intelligence is Qwen3.5-9B, at $0.11 per million tokens against $2.00.

Superseded. Claude Haiku 5.5 is Anthropic's newer model in this line, at $0.20 blended: see Claude Haiku 5.5 or compare the two.

Results 16 Nov 2026 · 146 results · 12 sources

Test Claude Haiku 4.5 on your prompt → It runs beside the models you could use instead, with every answer checked and costed.

Price per million tokens
$1.00 in · $5.00 out
Anthropic's list price, 10 Nov 2026 · compare prices$0.10 cached input · 50% off in batch
Speed
Not measured
Intelligence Index
16.9
Artificial Analysis, at thinking reasoning; 15.4–16.9 across 2 settings
Spring Prompt overall
36th of 36 score 8 of 100 · how it works
Spring Prompt Score
13 of 100 32nd of 32 · 5 of 5 benchmarks · our own tests only
Developer
Anthropic, based in the US
Weights not published
Released
15 Oct 2025
200,000 tokens of context

Where it stands among the models you could choose

Available models grouped by price, intelligence, speed and developer location. Retired models are excluded.

AmongIntelligence IndexSpring Prompt overallPrice, blendedSpeed
Models available today 105th of 185Claude Opus 5.5 35th of 35Claude Fable 5.2 $2.00 blended, 68th percentilecheapest: Mistral Nemo –
Models priced $1–3 per million tokens 29th of 34Muse Spark 1.3 8th of 8GPT-6.1 Terra sets this group –
Models with similar intelligence sets this group – $2.00 blended, 84th percentilecheapest: Qwen3.5-9B –
Models from developers based in the US 58th of 92Claude Opus 5.5 21st of 21Claude Fable 5.2 $2.00 blended, 48th percentilecheapest: Gemma 3 4B –

The name under each rank is the leader of that group; green is the top quarter of the group and red the bottom quarter, by rank. Price is shown as a percentile: the share of the group that costs less, so lower is cheaper. "Sets this group" marks the measure the group is defined by. Spring Prompt overall is our score out of 100 across every source (how it works). Price is the developer's list price (or the typical OpenRouter host where we haven't read one) for three input tokens to one output; intelligence and speed are from Artificial Analysis (data sourced from Artificial Analysis); "similar intelligence" means within 4 points on its Intelligence Index. Developer locations are where each company is based, not where a model is served. Groups under 3 models are not ranked.

Measured cost per task

API cost recorded in our benchmark runs, excluding judging.

TaskBenchmarkCost per task
Product listingCatalogBench≈ $0.006
DeckDeckBench≈ $0.046
Survey plan and analysisSurveyBench≈ $0.064
Game of blitz chessBulletBench≈ $0.032

Cheapest model that passes

On our benchmarks that define a pass for each task: what one passing answer cost in our run, failed attempts included, and whether a cheaper model passed.

  • CatalogBench · cost per 1,000 accepted listings
    Claude Haiku 4.5: $49.10 (21 of 168 attempts), 19 of 25. Cheapest: GPT-6 Luna at $0.70 (114 of 168 attempts). It passed at least as often.
  • DeckBench · cost per presentable deck
    Claude Haiku 4.5: no presentable decks (0 of 6 tasks). Cheapest: GPT-6.1 Terra at $0.13 (3 of 6 tasks).
  • SurveyBench · cost per sound survey task
    Claude Haiku 4.5: no sound survey tasks (0 of 9 tasks). Cheapest: GPT-6 Luna at $0.03 (3 of 9 tasks).

Cost per attempt divided by the share that passed, at the run date's prices, excluding judging. Each model at its cheapest setting per passing answer. With a handful of tasks, a pass or two either way is within chance. No estimate where a run had missing outputs, because their cost may not have been recorded.

How good is it, and for what?

Each use case ranks the available models its benchmarks measured, then those priced $1–3 per million tokens. The bar shows its percentile in that field, best to the right. Open a row for the results behind it.

Product listingsTurning a sparse product feed and photos into listings that can go live 29th of 32 available · 5th of 6 at its price
BenchmarkClaude Haiku 4.5Rank among availableBest available
CatalogBenchCatalogBench: reliably publish-ready 1.8% 29th of 32 GPT-6 Astra 75.0%
CatalogBenchCatalogBench (sales brief): reliably publish-ready 0.0% 27th of 32 GPT-6 Astra 69.6%
Decks from an analysisTurning a finished analysis into a deck you could present as it is 28th of 32 available · 4th of 6 at its price
BenchmarkClaude Haiku 4.5Rank among availableBest available
DeckBenchDeckBench: deck rating 658 28th of 32 Qwen4-Max 1,585
User surveysPlanning a user survey and reading its results without being misled 22nd of 23 available · 4th of 4 at its price
BenchmarkClaude Haiku 4.5Rank among availableBest available
SurveyBenchSurveyBench: analysis rating 569 22nd of 23 Claude Fable 5.2 1,641
Marketing planningPlanning a year of ad spend without overspending 31st of 32 available · 7th of 7 at its price
BenchmarkClaude Haiku 4.5Rank among availableBest available
ROASBenchROASBench: overall score 17.5 31st of 32 Claude Fable 5.2 61.3
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 98th of 159 available · 25th of 31 at its price
BenchmarkClaude Haiku 4.5Rank among availableBest available
Berkeley Function Calling Leaderboard (BFCL) V4BFCL: overall accuracy · reported 68.7% 4th of 36 Claude Opus 4.5 77.5%
Artificial AnalysisArtificial Analysis: Terminal-Bench 4.0 · reported 0.0% 67th of 107 Kimi K3.1 Pro 67.2%
Artificial AnalysisArtificial Analysis: Terminal-Bench 2.1 · reported 44.2% 67th of 107 Claude Fable 5.1 91.4%
Artificial AnalysisArtificial Analysis: τ-bench banking · reported 9.3% 75th of 104 Grok 4.6 50.7%
Vending-Bench 2Vending-Bench 2: money after a year · reported $458.89 46th of 53 GPT-6 Astra $15,515
  • tau2-bench: has not measured this model.
  • OpenHands Index: has not measured this model.
  • Microsoft STATE-Bench: has not measured this model.
Professional workReal tasks from banking, consulting and law, business documents and freelance projects Not measured
  • APEX-Agents: has not measured this model.
  • GDP.pdf: has not measured this model.
  • Remote Labor Index: has not measured this model.
Reasoning and knowledgeHard questions across science, maths and general knowledge 124th of 187 available · 32nd of 35 at its price
BenchmarkClaude Haiku 4.5Rank among availableBest available
Artificial AnalysisArtificial Analysis: Artificial Analysis Intelligence Index · reported 16.9 105th of 185 Claude Opus 5.5 57.6
Artificial AnalysisArtificial Analysis: Humanity's Last Exam · reported 10.4% 126th of 186 Claude Opus 5.5 61.4%
Artificial AnalysisArtificial Analysis: GPQA Diamond · reported 67.2% 121st of 168 GPT-6 Astra 96.3%
WritingWhat people prefer in blind comparisons, and judged writing quality 87th of 169 available · 25th of 28 at its price
BenchmarkClaude Haiku 4.5Rank among availableBest available
Arena (formerly LMArena)Arena Text: overall · reported 1,414 88th of 153 Gemini 4 Argon 1,525
UGI LeaderboardUGI: writing score · reported 52.2 54th of 124 Gemini 3.8 Flash 78.6
Sticking to the factsSummarising without inventing things, and factual answers 110th of 139 available · 25th of 26 at its price
BenchmarkClaude Haiku 4.5Rank among availableBest available
Vectara Hallucination LeaderboardVectara: hallucination rate · reported 9.8% 40th of 71 GPT-5.4 nano 3.1%
Arena (formerly LMArena)Arena Text factuality: overall · reported 1,418 87th of 109 Gemini 4 Argon 1,511
SimpleQA Verified (Epoch AI)SimpleQA Verified: correct answers · reported 13.2% 62nd of 68 GPT-6 Astra 75.6%
Speed under pressureGood decisions against a real clock (fast chess) 23rd of 35 available · 3rd of 7 at its price
BenchmarkClaude Haiku 4.5Rank among availableBest available
BulletBenchBulletBench Lightning 10+1: ladder Elo 33 16th of 18 Gemini 3.5 Flash-Lite 892
BulletBenchBulletBench Bullet 60s: ladder Elo 445 13th of 33 Gemini 3.8 Flash 1,068

Nearest alternatives

Against Claude Haiku 4.5 at $2.00 blended, Intelligence Index 16.9. "Similar intelligence" means within 4 points or better.

Where to run it

8 hosts on OpenRouter at the standard service tier, with a median of $2.10 blended per million tokens. Cheapest first.

HostQuantisationInputOutputBlended
Amazon Bedrocknot stated$1.00$5.00$2.00
Anthropicnot stated$1.00$5.00$2.00
Azurenot stated$1.00$5.00$2.00
Googlenot stated$1.00$5.00$2.00
Amazon Bedrocknot stated$1.10$5.50$2.20
Amazon Bedrocknot stated$1.10$5.50$2.20
Googlenot stated$1.10$5.50$2.20
Googlenot stated$1.10$5.50$2.20

Prices per million tokens from OpenRouter's public endpoint list, checked 10 Nov 2026.

Against the alternatives

The models you would most likely weigh it against: the leaders of the groups above.

Scroll sideways to see every alternative.

Claude Haiku 4.5Muse Spark 1.3
most intelligent at $1–3 per million tokens
Qwen3.5-9B
cheapest with similar intelligence
Claude Opus 5.5
most intelligent available today
GPT-6.1 Terra
highest-scoring overall at $1–3 per million tokens
Price per million tokens $2.00 $2.00$0.11$8.00$2.00
Tokens a second – 25411096–
Intelligence Index 16.9 48.113.357.6–
Spring Prompt overall 8 59–8280
Rank among available models, by use case
Product listings 29th of 32 17th–13th4th
Decks from an analysis 28th of 32 25th–7th8th
User surveys 22nd of 23 ––4th8th
Marketing planning 31st of 32 25th–9th2nd
Agents and tool use 98th of 159 13th111th3rd7th
Professional work – 8th–3rd–
Reasoning and knowledge 124th of 187 11th116th1st22nd
Writing 87th of 169 7th112th5th–
Sticking to the facts 110th of 139 11th–3rd–
Speed under pressure 23rd of 35 33rd––17th

Shaded figures are better than Claude Haiku 4.5's. A dash means no result.

Research and examples

Where it has been measured

29th most intelligent of the 34 models priced $1–3 per million tokens. Among models available today, in the bottom quarter for product listings, decks from an analysis, user surveys, marketing planning and sticking to the facts.

Artificial Analysis lists it but has not published a speed for it yet

Every result

146 published results from 12 sources, each in the source's own units, with the configuration that produced it.

BulletBench · measured by Spring Prompt · 15 results
MeasureValueRankConfigurationDated
BulletBench Blitz 3+2: cost per gameUS dollars, lower is better $0.0318 6 of 22 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
BulletBench Blitz 3+2: games lost on time% of games, lower is better 0.0% 1 of 22 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
BulletBench Blitz 3+2: invalid moves% of moves, lower is better 0.9% 22 of 22 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
BulletBench Blitz 3+2: ladder Eloladder Elo, higher is better 491
299–659
19 of 22 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
BulletBench Blitz 3+2: median move timemilliseconds, lower is better 1.3 s 4 of 22 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
BulletBench Bullet 60s: cost per gameUS dollars, lower is better $0.0163 48 of 71 Claude Haiku 4.5
provider default reasoning
1 Oct 2026
BulletBench Bullet 60s: games lost on time% of games, lower is better 25.0% 32 of 71 Claude Haiku 4.5
provider default reasoning
1 Oct 2026
BulletBench Bullet 60s: invalid moves% of moves, lower is better 0.3% 64 of 71 Claude Haiku 4.5
provider default reasoning
1 Oct 2026
BulletBench Bullet 60s: ladder Eloladder Elo, higher is better 445
135–626
42 of 71 Claude Haiku 4.5
provider default reasoning
1 Oct 2026
BulletBench Bullet 60s: median move timemilliseconds, lower is better 1.2 s 37 of 71 Claude Haiku 4.5
provider default reasoning
1 Oct 2026
BulletBench Lightning 10+1: cost per gameUS dollars, lower is better $0.0095 39 of 48 Claude Haiku 4.5
provider default reasoning
1 Oct 2026
BulletBench Lightning 10+1: games lost on time% of games, lower is better 83.3% 43 of 48 Claude Haiku 4.5
provider default reasoning
1 Oct 2026
BulletBench Lightning 10+1: invalid moves% of moves, lower is better 0.0% 1 of 48 Claude Haiku 4.5
provider default reasoning
1 Oct 2026
BulletBench Lightning 10+1: ladder Eloladder Elo, higher is better 33
0–248
45 of 48 Claude Haiku 4.5
provider default reasoning
1 Oct 2026
BulletBench Lightning 10+1: median move timemilliseconds, lower is better 1.3 s 42 of 48 Claude Haiku 4.5
provider default reasoning
1 Oct 2026

Not shown: Not your cost: prices are those charged on the run date.

CatalogBench · measured by Spring Prompt · 31 results
MeasureValueRankConfigurationDated
CatalogBench (sales brief): channel rules broken% of products, lower is better 4.2% 34 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): claims to check% of products, lower is better 44.6% 32 of 32 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): content quality% of checks, higher is better 93.9% 32 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): failed outputs% of products, lower is better 0.0% 1 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): missing UK information% of products, lower is better 0.0% 1 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): not findable% of products, lower is better 6.6% 28 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): publish-ready listings% of products, higher is better 3.0% 30 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): reliably publish-ready% of products, higher is better 0.0% 27 of 34 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): unsupported claims% of products, lower is better 91.7% 34 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): unsupported claimsclaims per product, lower is better 5.32 34 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): wrong attributes% of products, lower is better 51.2% 35 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench (sales brief): wrong category or variant% of products, lower is better 6.6% 33 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: channel compliance% of products, higher is better 100.0% 1 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: channel rules broken% of products, lower is better 0.0% 1 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: claims to check% of products, lower is better 22.6% 32 of 32 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: conflicts caught% of conflicts, higher is better 100.0% 1 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: content quality% of checks, higher is better 94.4% 31 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: cost per productUS dollars, lower is better $0.0061 13 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: decision accuracy% of decisions, higher is better 90.6% 33 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: failed outputs% of products, lower is better 0.0% 1 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: field accuracy% of missing fields, higher is better 93.7% 15 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: invented values% of filled values, lower is better 6.2% 35 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: missing UK information% of products, lower is better 0.0% 1 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: not findable% of products, lower is better 8.9% 30 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: publish-ready listings% of products, higher is better 12.5% 31 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: reliably publish-ready% of products, higher is better 1.8% 30 of 34 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: time per productseconds, lower is better 8.10 5 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: unsupported claims% of products, lower is better 73.2% 35 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: unsupported claimsclaims per product, lower is better 2.69 35 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: wrong attributes% of products, lower is better 48.2% 35 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026
CatalogBench: wrong category or variant% of products, lower is better 7.7% 33 of 35 Claude Haiku 4.5
provider default reasoning
30 Sep 2026

Not shown: Not your catalogue: invented products with generated images, so a model's result on your own feed can differ.

DeckBench · measured by Spring Prompt · 17 results
MeasureValueRankConfigurationDated
DeckBench: accurate decks% of tasks, higher is better 16.7% 25 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: caveat dropped% of tasks, lower is better 0.0% 1 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: clean layout% of tasks, higher is better 50.0% 14 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: cost per deckUS dollars, lower is better $0.0461 13 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: deck ratingrating, higher is better 658
564–734
30 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: design quality% of the maximum, higher is better 47.6% 29 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: draft figure quoted% of tasks, lower is better 50.0% 27 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: findings missing% of tasks, lower is better 16.7% 23 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: head-to-head win rate% of comparisons, higher is better 29.8% 30 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: layout defects% of tasks, lower is better 50.0% 14 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: misleading metric used% of tasks, lower is better 0.0% 1 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: presentable decks% of tasks, higher is better 0.0% 15 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: recommendation late or wrong% of tasks, lower is better 50.0% 26 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: slides needing work% of tasks, lower is better 66.7% 24 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: time per deckseconds, lower is better 51.7 5 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: unsupported claims% of tasks, lower is better 33.3% 24 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026
DeckBench: unsupported numbers% of tasks, lower is better 0.0% 1 of 34 Claude Haiku 4.5
provider default reasoning
6 Oct 2026

Not shown: Not your analysis or your brand: invented companies, six public tasks, one deck per model per task at the provider's default reasoning setting.

ROASBench · measured by Spring Prompt · 9 results
MeasureValueRankConfigurationDated
ROASBench: audience scorescore out of 100, higher is better 30.9 34 of 35 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
ROASBench: business scorescore out of 100, higher is better 7.3 34 of 35 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
ROASBench: consistency scorescore out of 100, higher is better 8.7 34 of 35 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
ROASBench: contribution profitsimulated US dollars, higher is better −$205,001 35 of 35 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
ROASBench: cost of a runUS dollars, lower is better $0.0298 5 of 35 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
ROASBench: months over budgetmonths of 12, lower is better 2 32 of 35 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
ROASBench: overall scorescore out of 100, higher is better 17.5 34 of 35 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
ROASBench: planning scorescore out of 100, higher is better 54.4 25 of 35 Claude Haiku 4.5
provider default reasoning
29 Sep 2026
ROASBench: return on ad spendprofit per $1 spent, higher is better −$0.21 35 of 35 Claude Haiku 4.5
provider default reasoning
29 Sep 2026

Not shown: Not real ad performance: the market is a deterministic simulation of one invented skincare brand.

SurveyBench · measured by Spring Prompt · 20 results
MeasureValueRankConfigurationDated
SurveyBench: analysis ratingrating, higher is better 569
498–621
23 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: analysis scorepoints of 100, higher is better 56.8 18 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: cost per taskUS dollars, lower is better $0.0635 10 of 24 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: head-to-head win rate% of comparisons, higher is better 24.6% 23 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: leading or double-barrelled questions% of tasks, lower is better 66.7% 21 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: mistaken next steps% of tasks, lower is better 55.6% 20 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: numbers right% of answers, higher is better 63.7% 24 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: objectives unanswerable% of tasks, lower is better 22.2% 21 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: precision ignored% of tasks, lower is better 100.0% 23 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: research scorepoints of 100, higher is better 64.9 18 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: sound research% of tasks, higher is better 0.0% 15 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: stakeholder's view presumed% of tasks, lower is better 22.2% 15 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: survey plan scorepoints of 100, higher is better 83.6 19 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: survey structure% of tasks, lower is better 0.0% 1 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: time per taskseconds, lower is better 92.6 7 of 24 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: traps handled% of traps, higher is better 62.3% 22 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: traps missed% of tasks, lower is better 88.9% 16 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: unsupported findings% of tasks, lower is better 100.0% 22 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: wrong numbers% of tasks, lower is better 100.0% 22 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026
SurveyBench: wrong recommendation% of tasks, lower is better 44.4% 17 of 25 Claude Haiku 4.5
provider default reasoning
8 Oct 2026

Not shown: Not your customers or your data: invented organisations with simulated respondents, nine public tasks, one run per model per task at the provider's default reasoning setting.

Arena (formerly LMArena) · reported by Arena (formerly LMArena) · 8 results
MeasureValueRankConfigurationDated
Arena Document: overallArena rating, higher is better 1,419
1,413–1,425
42 of 44 Claude Haiku 4.5 13 Sep 2026
Arena Text factuality: overallArena rating, higher is better 1,418
1,416–1,420
136 of 182 Claude Haiku 4.5 2 Oct 2026
Arena Text: business, management and financeArena rating, higher is better 1,420
1,415–1,424
126 of 406 Claude Haiku 4.5 2 Oct 2026
Arena Text: creative writingArena rating, higher is better 1,388
1,383–1,393
133 of 411 Claude Haiku 4.5 2 Oct 2026
Arena Text: expert promptsArena rating, higher is better 1,451
1,445–1,457
112 of 363 Claude Haiku 4.5 2 Oct 2026
Arena Text: instruction followingArena rating, higher is better 1,415
1,411–1,419
118 of 413 Claude Haiku 4.5 2 Oct 2026
Arena Text: overallArena rating, higher is better 1,414
1,411–1,416
142 of 413 Claude Haiku 4.5 2 Oct 2026
Arena Text: writing, literature and languageArena rating, higher is better 1,400
1,396–1,404
125 of 412 Claude Haiku 4.5 2 Oct 2026

Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Artificial Analysis · reported by Artificial Analysis · 19 results
MeasureValueRankConfigurationDated
Artificial Analysis: Artificial Analysis Coding Indexindex score, higher is better 43.9 114 of 187 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better 15.4 251 of 413 Claude Haiku 4.5 (no reasoning) 20 Oct 2026
Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better 16.9 236 of 413 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: GPQA Diamond% of questions, higher is better 64.6% 279 of 360 Claude Haiku 4.5 (no reasoning) 20 Oct 2026
Artificial Analysis: GPQA Diamond% of questions, higher is better 67.2% 261 of 360 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: Humanity's Last Exam% of questions, higher is better 4.2% 374 of 419 Claude Haiku 4.5 (no reasoning) 20 Oct 2026
Artificial Analysis: Humanity's Last Exam% of questions, higher is better 10.4% 275 of 419 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: IFBench% of instructions, higher is better 42.0% 191 of 288 Claude Haiku 4.5 (no reasoning) 20 Oct 2026
Artificial Analysis: IFBench% of instructions, higher is better 54.3% 123 of 288 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better 49.7% 283 of 407 Claude Haiku 4.5 (no reasoning) 20 Oct 2026
Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better 74.3% 157 of 407 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: SciCode% of problems, higher is better 42.2% 155 of 205 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: Terminal-Bench 2.1% of tasks, higher is better 44.2% 118 of 188 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: Terminal-Bench 4.0% of tasks, higher is better 0.0% 152 of 204 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: Terminal-Bench Hard% of tasks, higher is better 27.3% 117 of 282 Claude Haiku 4.5 (no reasoning) 20 Oct 2026
Artificial Analysis: Terminal-Bench Hard% of tasks, higher is better 27.3% 117 of 282 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: τ-bench banking% of tasks, higher is better 9.3% 130 of 175 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026
Artificial Analysis: τ²-bench telecom% of tasks, higher is better 32.5% 202 of 286 Claude Haiku 4.5 (no reasoning) 20 Oct 2026
Artificial Analysis: τ²-bench telecom% of tasks, higher is better 54.7% 155 of 286 Claude Haiku 4.5 (thinking reasoning) 20 Oct 2026

Not shown: Not your codebase or tools.

Berkeley Function Calling Leaderboard (BFCL) V4 · reported by Berkeley Function Calling Leaderboard (BFCL) V4 · 16 results
MeasureValueRankConfigurationDated
BFCL: irrelevance detection% correct, higher is better 85.1% 32 of 109 Claude Haiku 4.5 (bfcl fc) 16 Dec 2025
BFCL: irrelevance detection% correct, higher is better 95.3% 4 of 109 Claude Haiku 4.5 (bfcl prompt) 16 Dec 2025
BFCL: memory% correct, higher is better 54.4% 7 of 109 Claude Haiku 4.5 (bfcl fc) 16 Dec 2025
BFCL: memory% correct, higher is better 2.6% 99 of 109 Claude Haiku 4.5 (bfcl prompt) 16 Dec 2025
BFCL: multi-turn tasks% correct, higher is better 53.6% 14 of 109 Claude Haiku 4.5 (bfcl fc) 16 Dec 2025
BFCL: multi-turn tasks% correct, higher is better 1.8% 96 of 109 Claude Haiku 4.5 (bfcl prompt) 16 Dec 2025
BFCL: overall accuracy% correct, higher is better 68.7% 6 of 109 Claude Haiku 4.5 (bfcl fc) 16 Dec 2025
BFCL: overall accuracy% correct, higher is better 25.3% 87 of 109 Claude Haiku 4.5 (bfcl prompt) 16 Dec 2025
BFCL: relevance detection% correct, higher is better 62.5% 91 of 109 Claude Haiku 4.5 (bfcl fc) 16 Dec 2025
BFCL: relevance detection% correct, higher is better 31.3% 106 of 109 Claude Haiku 4.5 (bfcl prompt) 16 Dec 2025
BFCL: single-turn calls (curated)% correct, higher is better 86.5% 34 of 109 Claude Haiku 4.5 (bfcl fc) 16 Dec 2025
BFCL: single-turn calls (curated)% correct, higher is better 55.4% 98 of 109 Claude Haiku 4.5 (bfcl prompt) 16 Dec 2025
BFCL: single-turn calls (user-contributed)% correct, higher is better 78.7% 19 of 109 Claude Haiku 4.5 (bfcl fc) 16 Dec 2025
BFCL: single-turn calls (user-contributed)% correct, higher is better 52.5% 98 of 109 Claude Haiku 4.5 (bfcl prompt) 16 Dec 2025
BFCL: web search% correct, higher is better 83.5% 2 of 109 Claude Haiku 4.5 (bfcl fc) 16 Dec 2025
BFCL: web search% correct, higher is better 19.5% 45 of 109 Claude Haiku 4.5 (bfcl prompt) 16 Dec 2025

Not shown: Not general refusal or safety behaviour.

SimpleQA Verified (Epoch AI) · reported by SimpleQA Verified (Epoch AI) · 2 results
MeasureValueRankConfigurationDated
SimpleQA Verified: correct answers% of questions, higher is better 13.2% 77 of 85 Claude Haiku 4.5 10 Aug 2026
SimpleQA Verified: correct answers% of questions, higher is better 12.6% 79 of 85 Claude Haiku 4.5 (thinking-32k reasoning) 27 Aug 2026

Not shown: Not answers grounded in your documents; tests what the model remembers.

UGI Leaderboard · reported by UGI Leaderboard · 6 results
MeasureValueRankConfigurationDated
UGI: requested-length error% off the requested word count, lower is better 6.0% 49 of 379 Claude Haiku 4.5 (no reasoning) 15 Oct 2025
UGI: requested-length error% off the requested word count, lower is better 24.0% 225 of 379 Claude Haiku 4.5 (thinking reasoning) 15 Oct 2025
UGI: style adherencescore from 0 to 1, higher is better 0.35 187 of 379 Claude Haiku 4.5 (no reasoning) 15 Oct 2025
UGI: style adherencescore from 0 to 1, higher is better 0.39 44 of 379 Claude Haiku 4.5 (thinking reasoning) 15 Oct 2025
UGI: writing scorescore out of 100, higher is better 52.2 154 of 379 Claude Haiku 4.5 (no reasoning) 15 Oct 2025
UGI: writing scorescore out of 100, higher is better 49.1 171 of 379 Claude Haiku 4.5 (thinking reasoning) 15 Oct 2025

Not shown: Not other format limits such as character counts or bullet counts.

Vectara Hallucination Leaderboard · reported by Vectara Hallucination Leaderboard · 2 results
MeasureValueRankConfigurationDated
Vectara: answer rate% of documents, higher is better 99.5% 53 of 108 Claude Haiku 4.5 22 Sep 2026
Vectara: hallucination rate% of summaries, lower is better 9.8% 58 of 108 Claude Haiku 4.5 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.

Vending-Bench 2 · reported by Vending-Bench 2 · 1 result
MeasureValueRankConfigurationDated
Vending-Bench 2: money after a yearUS dollars, higher is better $458.89 54 of 63 Claude Haiku 4.5 1 Oct 2026

Not shown: Not a real business; one simulated market with set rules.

Embed a score badge

A badge for its Spring Prompt Score and each of our benchmarks that ranked Claude Haiku 4.5, from the current results. Paste the code into a page or README; it links back to where the result comes from.

Claude Haiku 4.5 is #32 of 32 on the Spring Prompt Score (13 of 100, from 5 of 5 of our benchmarks), measured by Spring Prompt Spring Prompt Score · 5 of 5 benchmarks
<a href="https://springprompt.com/score#anthropic-claude-haiku-4-5"><img src="https://springprompt.com/badge/score/anthropic-claude-haiku-4-5.svg" alt="Claude Haiku 4.5 is #32 of 32 on the Spring Prompt Score (13 of 100, from 5 of 5 of our benchmarks), measured by Spring Prompt" height="20"></a>
Claude Haiku 4.5 is #29 of 32 on CatalogBench (Reliably publish-ready, factual brief: 1.8%), measured by Spring Prompt CatalogBench · Reliably publish-ready, factual brief
<a href="https://springprompt.com/benchmarks/catalogbench?focus=anthropic-claude-haiku-4-5"><img src="https://springprompt.com/badge/catalogbench/anthropic-claude-haiku-4-5.svg" alt="Claude Haiku 4.5 is #29 of 32 on CatalogBench (Reliably publish-ready, factual brief: 1.8%), measured by Spring Prompt" height="20"></a>
Claude Haiku 4.5 is #29 of 33 on DeckBench (Deck rating: 658), measured by Spring Prompt DeckBench · Deck rating
<a href="https://springprompt.com/benchmarks/deckbench?focus=anthropic-claude-haiku-4-5"><img src="https://springprompt.com/badge/deckbench/anthropic-claude-haiku-4-5.svg" alt="Claude Haiku 4.5 is #29 of 33 on DeckBench (Deck rating: 658), measured by Spring Prompt" height="20"></a>
Claude Haiku 4.5 is #23 of 24 on SurveyBench (Analysis rating: 569), measured by Spring Prompt SurveyBench · Analysis rating
<a href="https://springprompt.com/benchmarks/surveybench?focus=anthropic-claude-haiku-4-5"><img src="https://springprompt.com/badge/surveybench/anthropic-claude-haiku-4-5.svg" alt="Claude Haiku 4.5 is #23 of 24 on SurveyBench (Analysis rating: 569), measured by Spring Prompt" height="20"></a>
Claude Haiku 4.5 is #32 of 33 on ROASBench (Score: 17.5), measured by Spring Prompt ROASBench · Score
<a href="https://springprompt.com/benchmarks/roasbench?focus=anthropic-claude-haiku-4-5"><img src="https://springprompt.com/badge/roasbench/anthropic-claude-haiku-4-5.svg" alt="Claude Haiku 4.5 is #32 of 33 on ROASBench (Score: 17.5), measured by Spring Prompt" height="20"></a>
Claude Haiku 4.5 is #42 of 45 on BulletBench (Lightning 10+1: 33), measured by Spring Prompt BulletBench · Lightning 10+1
<a href="https://springprompt.com/benchmarks/bulletbench?focus=anthropic-claude-haiku-4-5"><img src="https://springprompt.com/badge/bulletbench/anthropic-claude-haiku-4-5.svg" alt="Claude Haiku 4.5 is #42 of 45 on BulletBench (Lightning 10+1: 33), measured by Spring Prompt" height="20"></a>

Will Claude Haiku 4.5 work for your product?

Benchmarks aren't your task. Paste the prompt your product uses and run it beside the models you could use instead, with every answer checked and costed.

New here? Your email starts a free trial. No credit card needed.