Anthropic has launched Claude Haiku 5.5, a small model it says is designed for high-volume, cost-sensitive work such as summaries and classification. Its published API pricing starts at $0.10 per million input tokens and $0.50 per million output tokens, with higher rates for prompts over 100,000 tokens.
Anthropic Watch analysis
What happened
Anthropic describes Haiku 5.5 as the cheapest, fastest and most capable small model it has released. The company’s launch announcement says it is aimed at tasks where cost and volume matter. Anthropic’s Claude Code release notes list a one-million-token context window and the pricing above, with prompts over 100,000 tokens priced at $0.50 per million input tokens and $2.50 per million output tokens.
A separate Claude Code update had already added support for Haiku 5.5. This is the model launch itself, with Anthropic’s stated purpose and pricing, rather than another account of the SDK change.
Why it matters
For teams processing lots of text, the lower listed rate could make Haiku 5.5 worth testing for routine jobs such as classification and summarisation. The one-million-token context window is another concrete specification for developers weighing it against their existing options. Actual value will depend on how well it handles a team’s particular workload, not on the size of the context window alone.
Our read
This is a practical model release, not a claim that every task should now be handed to the smallest option on the menu. The price and context figures make Haiku 5.5 easy to shortlist; the sensible next step is to compare its quality and total cost on real workloads before routing a large queue through it.
What to watch
- How Haiku 5.5 performs on summaries and classification in independent evaluations.
- Whether its results justify the listed price for high-volume workloads.
- How developers use the one-million-token context window in practice.
Discussion spark: For high-volume tasks such as classification and summaries, would you choose a lower-cost small model by default, or pay more to reduce the risk of weaker results?
Sources and evidence
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.