Breaking

Claude Haiku 5.5 Now Available in Agent Arena Benchmark Platform

October 7, 2026 at 14:52 EDT

Foundation Models
AI Agents

Anthropic's fastest and cheapest Claude model is now open for real-world agentic evaluation, with performance benchmarks showing substantial gains over its predecessor. Claude Haiku 5.5 launched on October 7, 2026 with a focus on high-volume, latency-sensitive tasks, and is now available for testing through Agent Arena, a leaderboard that ranks AI models on millions of real-world, long-horizon agentic workflows from a global community of users.

OCTOBER 7, 2026 · ANTHROPIC

Haiku 5.5 Enters Agent Arena with 72× Computer-Use Capability

Anthropic's fastest, most cost-efficient Claude model now benchmarked on real-world agentic workflows—72.4% on OSWorld 2.1, up from 15.7%, and 75–90% cheaper than the prior generation.

72.4%

OSWorld 2.1 Score

90%

Cheaper than Haiku 4.5

92.8%

HubSpot CRM Test

OSWorld 2.1 Improvement

15.7%

Haiku 4.5

72.4%

Haiku 5.5

4.6× improvement in computer-use capability

API Cost Reduction

Cost per Million Tokens

Input: $0.10 (input)
Output: $0.50 (output)

75–90%

Cheaper than 4.5

Agent Arena Evaluation

Tested on millions of real-world agentic tasks with access to web search, filesystem, and terminal tools. Scores pending as data accumulates from user sessions. Other Claude models ranked: Fable 5.1 (#1), Opus 5.5 (#2), Sonnet 5.5 (#3).

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.