What happened
On October 7, 2026, Anthropic released Claude Haiku 5.5 (claude-haiku-5-5). It is the first Haiku with an adjustable effort setting, a 1-million-token context window and a tokenizer matching Sonnet/Opus 5.5.
Reported pricing and performance
- $0.10 input / $0.50 output per million tokens ($0.50 / $2.50 above 100K tokens) — about 75% less than Haiku 4.5 on average;
- 72.4% on OSWorld 2.1 (offline), versus 15.7% for Haiku 4.5 (Sonnet 5.5: 83.9%);
- 39.2% on Terminal-Bench 4.0 (Sonnet 5.5: 70.6%).
The same day, Sonnet 5.5's cache-read price was halved ($0.20 → $0.10), and Max/Team subscribers began receiving monthly API credits from $100 to $500. Claude Code uses Haiku 5.5 as its default Haiku from version 2.1.293.
Why it matters
Much agent work (summarize, classify, subagents) does not need the most expensive model. When that layer's cost drops by an order of magnitude, the "Sonnet/Opus conducts + Haiku executes" architecture changes the bill.
What to do now
- Put Haiku 5.5 into summarization and subagent steps and measure cost and quality.
- Recalculate tokens: the tokenizer changed, so renaming the model alone misleads cost estimates.
- Pair it with Sonnet 5.5's cheaper cache on long flows.