Claude Haiku 5.5 becomes the default Haiku with 1M context at $0.10/$0.50 per Mtok

Anthropic's newest small model, Claude Haiku 5.5, arrived through a Claude Code release and is now the default when developers request Haiku on the API (model ID claude-haiku-5-5). The headline numbers are aggressive. It has a 1-million-token context window, priced at $0.10 per million input tokens and $0.50 per million output tokens. Prompts over 100K tokens move to a higher tier of $0.50/$2.50 per Mtok, so long-context use is still five times more expensive than short calls.

The two-tier pricing reflects how small models are actually used in agent systems. Most Haiku calls are short: classification, tool routing, summarizing a tool result, or acting as a subagent in a larger Opus-led workflow. Pricing that path at $0.10/$0.50 makes it cheap to fan out many subagents. The 1M window remains available for jobs like scanning a codebase or a long document, but at a premium. Promoting it to the default alias means existing integrations that point at "Haiku" pick it up automatically, which is convenient but can change behavior without a code change.

Competitively, Haiku 5.5 is Anthropic's response to the low-cost pressure from DeepSeek's V4.1-Flash and Google's Flash-tier models, and it pairs naturally with Claude Code's new client-side auto model selection. Builders are already discussing it as a default for high-volume agent subagents. It arrives the same week Accenture capped central Claude Enterprise spend at $29 per employee, so the market for cheaper Claude tokens is plain to see.

Caveats: Anthropic has not highlighted benchmark comparisons alongside the release. Teams relying on the floating Haiku alias should regression-test before production traffic shifts. The 100K-token threshold also makes cost forecasting less simple for retrieval-heavy workloads.