None
None
Long Story Video Skill
Turn your idea into 10s–10min video—AI writes scripts, prompts, and generates footage automatically.
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
Feedback
freeTrialImage.bannerPity
freeTrialImage.upgradeUnlock
- ✓freeTrialImage.benefitHd
- ✓freeTrialImage.benefitWatermark
- ✓freeTrialImage.benefitUnlimited
Haiku 5.5 Benchmark
See how the Haiku 5.5 Benchmark performs: measured task results, per-token costs and context limits, all tied to their published sources.
All Tools
Discover our comprehensive AI-powered animation toolkit
MiniMax H3
MiniMax H3 AI Video Generator
Seedance 2.5
The Future of AI Video Is Here.
Seedance 2.0
The Future of AI Video Is Here.
Veo3.1
Create Stunning Videos with Veo3.1
FLUX 3 Video Generator
Kling 3.0
Next-Gen AI Video Generator
Grok Video Generator
Create Videos from Text or Images with AI
MiniMax H3 video generator
MiniMax H3 AI Video Generator
Haiku 5.5 Benchmark: How Each Figure Is Sourced
Every released evaluation score, cost band and reasoning-effort price stays paired with the setup it ran on and the date it was pulled.
- Scores That Cannot Be Averaged TogetherElo ratings, partial credit, coding passes and accuracy rates sit on separate scales, so blending them into one number misleads.
- Economics of a Small Model at VolumeDesigned for extraction, classification and summarization jobs that run thousands of times a day.
- Two Footnotes That Rewrite the TablesLow pricing applies only up to 100K tokens, and the 72.4% computer-use figure is partial credit rather than a pass rate.
Putting the Haiku 5.5 Benchmark to Work
Three moves that convert released Claude Haiku 5.5 evaluation figures into a call about your own workload.
Haiku 5.5 Benchmark: What the Specs and Scores Show
Released specs, pricing bands, coding outcomes, computer-use results and third-party index readings for Claude Haiku 5.5.
Context Window and Output Ceiling
Text and images go in, text comes out, with adaptive thinking enabled. A 1M-token context and a 128,000-token output cap far exceed Haiku 4.5's 200,000 and 64,000.
Pricing Splits at Two Prompt Lengths
Past 100,000 prompt tokens, input, output and caching each move to a tier five times the cheap rate. OpenAI's base Standard band reaches 272,000 input tokens.
Terminal-Bench 4.0 Finished Tasks
The 39.2% launch result tops Luna by a clear margin yet trails the published Sol rows. Runs allow an eight-hour agent timeout, and the 2.1 or 3.0 figures come from an earlier experiment.
FrontierCode Mergeability and Reasoning Spend
Scoring weighs 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 xhigh to max doubles cost for half a point.
OSWorld 2.1: Partial Credit vs. a Clean Pass
The run covers 82 offline tasks, five attempts each, 1080p screenshots and a 500-step cap. When a workflow has to finish, the 37.1% strict rate is the figure to trust.
Third-Party Index Scores and Tool Gains
Artificial Analysis lists Haiku across all five efforts on Intelligence Index v4.3.2 with rounded task costs. HLE shows 45.9% without tools and 57.4% with them — uplift that comes from the surrounding system.
Haiku 5.5 Benchmark: Questions Answered
Answers on Claude Haiku 5.5 evaluation results, token pricing and the caps that shape real usage.
How does Claude Haiku 5.5 stack up against GPT-6 Luna and GPT-6 Sol?
Across knowledge tasks, screenshot work and narrow coding, Haiku and Luna trade wins. Sol and Sol 6.1 charge more and justify it on coding, yet Sonnet 5.5 still tops most rows.
Why shouldn't the 72.4% computer-use figure be read as a success rate?
Require every checkpoint to pass and the same run drops to 37.1%. An agent may find and edit the correct file, then save it in the wrong format or halt before the final step. Always add an end-state check.
At what point does Claude Haiku 5.5 move into the higher price band?
Prompts of up to 100,000 tokens qualify for the lowest rates; beyond that every category costs five times as much. The 90%, 50% and roughly 75% savings figures compare different baselines.
Is local hosting possible for Claude Haiku 5.5, and what GPU memory would it take?
The released specs disclose no parameter count, active-parameter number, layer count or training compute. 'Small' describes a product tier rather than an architecture, so any local GPU estimate lacks a basis.
Does raising reasoning effort always deliver better output?
No. Sol 6.1 scores higher at medium than at max, and Haiku's 0.2-point lead over Sonnet at max is too slim to count as a win. More effort also means more tokens, more latency and more chances to drift off task.
Why do identical models post different numbers across tables?
Sonnet shows 70.6% in Anthropic's launch table but 61.8% on the public leaderboard snapshot, where Haiku had no visible row. Artificial Analysis removed its 'Default Fallback' label, while Anthropic's migration guide still states no server-side fallback exists.
Put the Haiku 5.5 Benchmark Up Against Your Own Tasks
Choose a single workload, fix the reasoning effort, and see whether Claude Haiku 5.5 clears the quality bar you already hold.