BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9
AnthropicCompare with another model

Claude 3 Haiku

Claude 3 Haiku is an Anthropic proprietary model, released 7 Mar 2024. Its best configuration ranks #738 of 785 on the BenchLeader Index at 37.0 ±4.0, in the lower half. The ± is the point: 175 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in human preference (41) and lowest in multimodal (19). At $0.500 per million tokens blended it is cheaper than most ranked models. Last measured 22 Oct 2026.

Blended price
$0.500/M
$0.250 in · $1.25 out
Output speed
–
First answer
–
Context
200k
Full answer
–
Cost per run
–
Released
7 Mar 2024
How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.
  1. Overall index37
  2. Reasoning36
  3. Coding32
  4. Agents & tools34
  5. Maths33
  6. Knowledge40
  7. Instruction following39
  8. Human preference41
  9. Multimodal19
  10. Long context38
  11. Composite36

Versions

Anthropic has shipped 6 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

ModelReleasedIndexRank
Claude Haiku 5.5xhigh7 Oct 202661.5#95
Claude Haiku 4.5no reasoning15 Oct 202548.5#424
Claude 3.5 Haiku4 Nov 202439.7#675
Claude 3 Haikuthis page7 Mar 202437.0#738
Claude Haiku 3–––
Claude Haiku 4thinking–––

How its index has moved

8 Sept 2026 to 21 Oct 2026
323742

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

Benchmarknot statedSourceTrend
GPQA Diamond36.3%#257Epoch AI Benchmarking Hub
Arena Hard Prompts1266#284Arena (LMArena)
GPQA Diamond (AA)not in index37.4%#487Artificial Analysis
Humanity's Last Exam (AA)not in index4.1%#544Artificial Analysis
CritPt0.0%#315Artificial Analysis
LMCAnot in index8.8%#182Epoch AI Benchmarking Hub
DTBenchnot in index50.1%#213Epoch AI Benchmarking Hub
ForecastBenchnot in index54.8%#83ForecastBench
WinoGrandenot in index74.2%#35Epoch AI Benchmarking Hub

Coding

Benchmarknot statedSourceTrend
WeirdML9.8%#151WeirdML
Arena Coding1301#278Arena (LMArena)
SWE-bench Verified (any scaffold)not in index40.6%#45SWE-bench
CadEvalnot in index12.0%#15Epoch AI Benchmarking Hub

Agents & tools

Benchmarknot statedSourceTrend
Terminal-Bench Hard0.8%#349Artificial Analysis
τ²-Bench Telecom (AA)not in index21.1%#326Artificial Analysis

Maths

Benchmarknot statedSourceTrend
OTIS Mock AIME1.8%#263Epoch AI Benchmarking Hub
MATH Level 514.9%#84Epoch AI Benchmarking Hub
Arena Maths1232#276Arena (LMArena)

Knowledge

Benchmarknot statedSourceTrend
AA-Omniscience-48.6#380Artificial Analysis
MMLUnot in index73.8%#42Epoch AI Benchmarking Hub
AA-Omniscience: accuracynot in index17.6%#361Artificial Analysis
AA-Omniscience: non-hallucinationnot in index19.6%#285Artificial Analysis

Instruction following

Benchmarknot statedSourceTrend
IFBench36.1%#307Artificial Analysis
Arena Instruction Followingnot in index1245#282Arena (LMArena)

Human preference

Benchmarknot statedSourceTrend
Arena Text1262#284Arena (LMArena)
Arena Creative Writingnot in index1215#283Arena (LMArena)
Arena Multi-turnnot in index1245#279Arena (LMArena)
Arena Longer Queriesnot in index1268#279Arena (LMArena)

Multimodal

Benchmarknot statedSourceTrend
Arena Vision950#151Arena (LMArena)
MMMU-Pro30.8%#274Artificial Analysis
MMMU (validation)50.2%#44MMMU
ScienceQAnot in index72.0%#11Epoch AI Benchmarking Hub

Long context

Benchmarknot statedSourceTrend
AA-LCR27.7%#404Artificial Analysis

Composite

Benchmarknot statedSourceTrend
Epoch Capabilities Indexnot in index118.8#179Epoch AI Benchmarking Hub
AA Intelligence Index v4.3.25.6#571Artificial Analysis

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.

WorkloadTokens in / outCostWith cachingTime
Chat reply400 / 300$0.0005––
Summarize a 30-page report12,000 / 600$0.0038––
Code edit6,000 / 1,500$0.0034––
Agentic coding session60,000 / 4,000$0.020––
Structured extraction2,000 / 200$0.0008––

More from Anthropic

New this month

See also

Best under $2/MBest for codingAll Anthropic models

Questions people ask

What are Claude 3 Haiku's benchmark scores?
As of 25 Oct 2026, Claude 3 Haiku scores 37.0 ±4.0 on the BenchLeader Index, strongest in human preference (41) and weakest in multimodal (19). That rests on 17 independent benchmark results, and every score on this page links to its publisher.
How does Claude 3 Haiku rank?
Claude 3 Haiku is #738 of 785 ranked configurations on 25 Oct 2026. 175 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
How much does Claude 3 Haiku cost per 1M tokens?
As of 25 Oct 2026, Claude 3 Haiku costs $0.250 per 1M input tokens and $1.25 per 1M output tokens, $0.500 blended (three input tokens per output token).
When was Claude 3 Haiku released?
Anthropic released Claude 3 Haiku on 7 Mar 2024.

Data as of 25 Oct 2026. Compare with another model.

Cite as: BenchLeader, “Claude 3 Haiku: benchmarks, pricing, speed and rank”, data as of 25 Oct 2026, https://www.benchleader.com/models/claude-3-haiku.