AI Research Atlas

Claude Haiku 4.5

Anthropic · 15 October 2025

Claude Haiku 4.5 matches Sonnet 4 coding at one-third the cost and over twice the speed. It scores 73.3% on SWE-bench Verified at $1/$5.

Fastest Claude, more than twice as fast as Sonnet 4 and four to five times faster than Sonnet 4.5, with about 90% of Sonnet 4.5's agentic coding performance. It surpasses Sonnet 4 at some computer-use tasks. Released under ASL-2, unlike the ASL-3 Sonnet 4.5.

Date
Wednesday, 15 October 2025
Lab
Anthropic
Kind
model
Access
closed API
Price
$1/$5 per M tokens in/out

Figures

MeasureValueMeasured by
SWE-bench Verified73.3%
Claude Haiku 4.5
company
Terminal-Bench40.21%
41.75% with 32K thinking
company

Still Anthropic's current Haiku as of 2026-10-04; a Haiku 5.5 was promised 'in the coming weeks' on the Opus 5.5 page (2026-09-22) and had not been found released. Retirement not sooner than 2026-10-15 per models overview.

Sources

  1. www.anthropic.com/news/claude-haiku-4-5

This record was checked against its sources on 6 October 2026. How we check

Related