Claude Haiku 5.5: Anthropic’s Cheapest, Fastest Small Model Yet

Anthropic just refreshed the budget end of the Claude stack. On October 7, 2026, the lab introduced Claude Haiku 5.5 as its cheapest, fastest and most capable small model yet — aimed at high-volume work, snappy agents and computer-use tasks that used to feel too expensive on larger models.

What it is for

• Quick, repetitive jobs: summaries, compaction, database queries, classification
• Speed-sensitive flows such as live customer support and browser use
• Coding subagents alongside Opus 5.5 or Sonnet 5.5

Haiku 5.5 is the first Haiku-class Claude with an adjustable effort setting, so teams can trade cost for intelligence per request. On the Claude Platform the model id is claude-haiku-5-5; it is also listed on AWS, Google Cloud and Microsoft Azure.

Price and value

Anthropic says Haiku 5.5 costs about 75% less on average than Haiku 4.5. For prompts up to 100,000 tokens (most of prior Haiku traffic), list pricing is $0.10 per million input tokens and $0.50 per million output; over 100k those rise to $0.50 / $2.50. Cache reads and writes are also far cheaper than Haiku 4.5.

The same day, Anthropic halved Sonnet 5.5 cache reads to $0.10/M (about 20% cheaper agentic work) and said Max/Team subscribers will get monthly API credits — Max 5x $100, Max 20x $200, Team up to $500 pooled — to experiment on the Claude Platform.

Early scores and quotes

Anthropic’s tables put Haiku 5.5 far ahead of Haiku 4.5 on OSWorld computer use (72.4% offline subset vs 15.7%) and agentic coding benches. Early customers quoted in the post report lower latency and better hit rates on CRM, document Q&A and teammate-style agents (Asana, HubSpot, AlphaSense, Box, Devin).

Rocks take

Cheap, fast subagents are how local and cloud agent stacks scale. If you already wire Claude into Home Assistant MCP tools or coding CLIs, Haiku 5.5 is the model to re-benchmark first — then decide what still needs Sonnet.