Anthropic on October 7 introduced Claude Haiku 5.5, describing it as the cheapest, fastest and most capable small model the company has released. The launch, detailed in a post on Anthropic's website, lands almost exactly one year after Haiku 4.5 shipped on October 15, 2025, and signals that Anthropic now treats price, not just raw capability, as a competitive front in its race against OpenAI and Google.
The new model costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, according to pricing details reported by The New Stack, versus $1 and $5 for the same measures under Haiku 4.5 — a roughly 75% cut that Anthropic says reflects typical usage patterns. Cached prompt reads fall to as little as $0.01 per million tokens. Above the 100,000-token threshold, pricing rises to $0.50 and $2.50 per million tokens.
Despite the lower price, Anthropic's own benchmark figures show Haiku 5.5 well ahead of its predecessor: 72.4% versus 15.7% on the OSWorld computer-use test, 45.9% versus 10.2% on Humanity's Last Exam without external tools, and 39.2% versus 0% on Terminal-Bench, a measure of agentic coding tasks. Independent benchmarking firm Artificial Analysis put Haiku 5.5's score on its Intelligence Index at 43, which it described as up 26 points from the prior Haiku release a year earlier.
Pricing as the New Battleground
Haiku 5.5 is also the first Haiku-class model to offer an adjustable "effort" setting, letting developers trade intelligence for speed and cost on a sliding scale from Low to Max. Anthropic has built much of its recent strategy around pairing small, cheap models like Haiku with its larger Sonnet and Opus models, using Haiku instances as subagents that larger models delegate repetitive subtasks to — a design meant to cut the cost of running complex, multi-step AI agents at scale.
Early customer testimonials in Anthropic's announcement, including from Asana, HubSpot and Box, emphasized the model's speed for high-volume tasks such as summarization and classification. Not every account has been flattering: Neowin reported that despite the aggressive pricing, Haiku 5.5 tends to consume more tokens per task than its predecessor, meaning real-world savings may be smaller than the sticker price suggests for some workloads.
Anthropic is simultaneously cutting prices elsewhere in its lineup, halving the cache-read price for Sonnet 5.5 to $0.10 per million tokens, a sign that cost competition among frontier AI providers is intensifying even as capability gaps between them narrow. The rollout is immediate across Anthropic's API and Claude apps, with monthly API credits bundled into existing Max and Team subscription tiers.
