Haiku 5.5 Benchmark: Model Profile
A directory entry for Claude Haiku 5.5, summarizing benchmark results, provider details, model family, input/output modes and API access notes.
None

None

None

Long Story Video Skill

Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.

Ads Video Skill

Ads Video Skill

Generate professional ads and sales videos—AI auto-generates scripts, prompts, and footage.

3D Science Explainer Video Skill

Convert scientific concepts into stunning 3D explain animations

AI Video Prompt Generator

Feedback

freeTrialImage.bannerPity

freeTrialImage.upgradeUnlock

  • ✓freeTrialImage.benefitHd
  • ✓freeTrialImage.benefitWatermark
  • ✓freeTrialImage.benefitUnlimited

Haiku 5.5 Benchmark

See how the Haiku 5.5 Benchmark performs in public evaluations, what each pricing tier costs, and which documented limits shape real-world use.

All Tools

Discover our comprehensive AI-powered animation toolkit

A Practical Lens on the Haiku 5.5 Benchmark

Reported evaluation scores, tiered pricing and effort-level costs, each documented with its test setup and access date.

  • How to Interpret the Reported Scores
    Elo, partial-credit, coding and accuracy metrics sit on different scales, so averaging them would be misleading.
  • Cost Efficiency for High-Volume Tasks
    Designed for extraction, classification and summarization jobs that run at scale.
  • Two Limits That Reshape the Score Tables
    Low rates end at 100K tokens, and the 72.4% computer-use result is partial credit rather than a strict pass rate.

Using the Haiku 5.5 Benchmark in Three Steps

A three-step method for translating published Haiku 5.5 Benchmark figures into a real workload decision.

Haiku 5.5 Benchmark: What the Data Shows

Documented specs, pricing tiers, coding outcomes, computer-use results and third-party index data for Claude Haiku 5.5.

Context and Output Capacity

The model accepts text and images and returns text with adaptive thinking enabled. A 1M-token context window and 128K-token output cap far exceed Haiku 4.5's 200K and 64K.

Pricing Split by Prompt Length

Beyond 100K prompt tokens, input, output and caching move to a tier priced five times higher. OpenAI's base Standard tier extends to 272K input tokens.

Terminal-Bench 4.0 Completion Rate

Haiku's 39.2% launch result clearly leads Luna but trails the published Sol entries. Task runs allow an eight-hour agent timeout; 2.1 and 3.0 figures come from an earlier experiment.

FrontierCode Mergeability and Cost per Effort

Scoring includes correctness, tests, scope, style and repository conventions. Haiku reaches 46.4% at max effort and 41.6% at medium for $0.13 per rollout, while moving from xhigh to max doubles cost for half a point.

OSWorld 2.1: Partial vs. Strict Success

The run uses 82 offline tasks, five attempts each, 1080p screenshots and a 500-step limit. For workflows that must complete, the 37.1% strict pass rate is the realistic figure.

Third-Party Index Results and Tool Gains

Artificial Analysis lists Haiku across all five effort levels on Intelligence Index v4.3.2 with rounded task costs. HLE scores 45.9% without tools and 57.4% with them, a lift tied to the surrounding system.

FAQ

Haiku 5.5 Benchmark: Common Questions

Answers to frequent questions on Claude Haiku 5.5 Benchmark results, pricing tiers and operational limits.

1

How does Haiku 5.5 stack up against GPT-6 Luna and GPT-6 Sol?

Haiku matches Luna closely on knowledge work, screenshot tasks and focused coding. Sol and Sol 6.1 charge more and justify it on coding, while Sonnet 5.5 remains ahead on most rows.

2

Why doesn't the 72.4% computer-use score equal a success rate?

When every checkpoint must be satisfied, the same run reports just 37.1%. An agent may find and edit the correct file, then save it in the wrong format or halt before the final action. Add an end-state check.

3

When does Haiku 5.5 move into the higher price tier?

The lowest rates apply only to prompts of up to 100K tokens; beyond that, every category costs five times as much. Anthropic's 90%, 50% and roughly 75% savings figures use different baselines.

4

Can Haiku 5.5 run locally, and what GPU memory would it require?

The published specs disclose no parameter count, active-parameter figure, layer count or training compute. 'Small' describes a product tier rather than an architecture, so local GPU estimates lack a basis.

5

Does higher reasoning effort always produce better results?

No. Sol 6.1 scores higher at medium than at max, and Haiku's 0.2-point lead over Sonnet at max is too slight to count as a win. More effort adds tokens, latency and opportunities to drift off task.

6

Why do identical models show different figures across tables?

Sonnet shows 70.6% in Anthropic's launch table and 61.8% in the public leaderboard snapshot, where Haiku had no visible row. Artificial Analysis removed its 'Default Fallback' label, but Anthropic's migration guide still states that no server-side fallback exists.

Put the Haiku 5.5 Benchmark to Work on Your Own Tasks

Choose a single workload, set the effort level, and measure Haiku 5.5 against the quality bar your team requires.