Anthropic Releases Claude Haiku 5.5 With Reasoning Level Control
A cheap-tier model that beats a rival's scores and undercuts the previous Haiku's API prices reframes the low-cost tier as the place Anthropic now pitches as a companion to its larger models rather than a fallback.
Reporting from 1 source: GIGAZINE.
Anthropic released Claude Haiku 5.5 on October 7, 2026, the Haiku tier's first update in a year. It scores higher than Claude Haiku 4.5 on benchmarks and surpasses GPT-6 Luna. Artificial Analysis measured its Intelligence Index v4.3.2 score at 43, up from 17 for Haiku 4.5. It is available on all plans, including free, and via API.
Haiku 5.5 is the first model in the Haiku series to support reasoning-level adjustment, which lets users trade cost for score.
On OSWorld 2.1, which measures AI PC operation performance, Anthropic's cost-per-task chart puts Haiku 5.5 above GPT-6 Luna in performance per cost. At the High reasoning setting it reaches a higher score at lower cost than Claude Sonnet 5.5 at Low.
Anthropic proposes running it alongside Claude Opus 5.5 or Claude Sonnet 5.5, and a demo shows the Opus plus Haiku combination processing faster and cheaper than Opus alone. Alongside the release, the cache read price for Claude Sonnet 5.5 was halved.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.