Anthropic launches Claude Haiku 5.5, its fastest small model

Anthropic unveiled Claude Haiku 5.5 on October 8, 2026, describing it as the cheapest, fastest and most capable small model released to date. The model costs 75 percent less to run than Haiku 4.5 while its Computer Use score jumps from 15.7 percent to 72.4 percent. Haiku 5.5 targets high‑volume, low‑cost tasks such as summarisation, context trimming, database retrieval and routing requests. It also serves as a subagent that offloads routine work from Opus 5.5 and Sonnet 5.5, enabling real‑time customer service and browser automation. Benchmarks show Haiku 5.5 scoring 1,620 on GDPval‑AA v2.1, more than double Haiku 4.5’s 735 and ahead of OpenAI’s GPT‑6 Luna’s 1,437. On AA‑Briefcase v1.1 it reaches 1,578 points versus 614 for Haiku 4.5 and 1,336 for GPT‑6 Luna.

Its Computer Use performance on OSWorld 2.1 rises to 72.4 percent, far surpassing Haiku 4.5’s 15.7 percent and closing the gap with Sonnet 5.5’s 83.9 percent. On Humanity’s Last Exam it reaches 45.9 percent without tools and 57.4 percent with tools, up from 10.2 percent and 18.7 percent respectively. In agentic coding on Terminal‑Bench 4.0 it achieves 39.2 percent, the first non‑zero result for a Haiku model, while GPT‑6 Luna manages 16.4 percent and Sonnet 5.5 70.6 percent. Anthropic notes that Haiku 5.5 still lags behind Sonnet 5.5 on Terminal‑Bench 4.0, so complex coding should use the larger model. The new model is the first Haiku to expose an “Effort” knob that lets users trade cost for reasoning depth, and a pricing graph shows accuracy versus expense across OSWorld, GDPval‑AA and Humanity’s Last Exam.

Early customers including Asana, HubSpot, AlphaSense, Box, Rogo and Cognition report speed gains of up to 2.5 times and a 30 percent reduction in per‑task latency. Token pricing drops tenfold for 90 percent of prompts, with input costing 0.10 dollars per million tokens for prompts up to 100,000 tokens and 0.50 dollars for longer prompts, while cache read and write prices are halved. Anthropic also cuts Sonnet 5.5 cache read prices by 50 percent and grants monthly API credits to Max and Team members.

Reported by Techsauce.