October 7, 2026
Subagent tasks get cheaper: Haiku 5.5 costs 75% less than Haiku 4.5 on average
Anthropic estimates Haiku 5.5's average operating cost at 75% less than Haiku 4.5's: the model has been available since Oct 7.
Addy Osmani
@addyosmani
Claude Haiku 5.5 is here 🎉 It's an affordable, fast and compact model with a broad range of capabilities. It costs roughly 75% less to use than Haiku 4.5, and you can adjust the model's effort. Our team really enjoys using it as a subagent alongside Opus 5.5.
Haiku 5.5 is now available in the Claude Platform and Claude Code. On average, it costs around 75% less to run than Haiku 4.5. It pairs well with Opus 5.5 or Sonnet 5.5 as a subagent. Use it for high-volume, cost-sensitive tasks like summaries, compactions, or database queries.
· 5.9K views
Haiku 5.5 handles high-volume tasks alongside Opus 5.5 or Sonnet 5.5: summarizing texts, compressing context and querying databases.
For AI coding agents, this means dividing work between models. Anthropic recommends Haiku for tasks where the cost of each run matters, and Sonnet or Opus for complex programming.
Pricing depends on volume. According to Anthropic's Oct 7 pricing, requests with up to and including 100,000 input tokens cost 90% less per token than Haiku 4.5. Above that threshold, the reduction is 50%.
| Input request | Per 1 million input tokens | Per 1 million output tokens | | --- | --- | --- | | Up to and including 100,000 tokens | $0.10 | $0.50 | | Over 100,000 tokens | $0.50 | $2.50 |
The new tokenizer counts roughly 30% more tokens for the same text than Haiku 4.5. When migrating, Anthropic requires recalculating prompts, `max_tokens` and costs. The model accepts up to 1 million tokens of context and generates up to 128,000 tokens per response, while the Batch API cuts input and output prices by another 50%.
Migration with a skill. For projects that call the Claude API, Claude Code provides the command `/claude-api migrate this project to claude-haiku-5-5`. The built-in skill updates the model identifier and incompatible parameters.
The Claude API uses `claude-haiku-5-5`, while Amazon Bedrock uses `anthropic.claude-haiku-5-5`. Haiku 5.5 is also available in Google Cloud and Microsoft Foundry.
Replace the previous `thinking: {"type":"enabled","budget_tokens":N}` with `thinking: {"type":"adaptive"}`. Remove `temperature`, `top_p` and `top_k`: the API returns HTTP 400 with the old thinking configuration or these parameters.
The model's effort is set through `output_config.effort`. The default is `medium`, which Anthropic recommends for agentic coding. For short, simple tasks, the company recommends `low`.
Available values:
- `low` - `medium` - `high` - `xhigh` - `max`
In Anthropic's Oct 7 tests, Haiku 5.5 improved on Haiku 4.5's Terminal-Bench 4.0 score. This benchmark tests an agent's ability to handle tasks in a terminal.
| Model | Terminal-Bench 4.0 | | --- | --- | | Haiku 4.5 | 0.0% | | Haiku 5.5 | 39.2% | | Sonnet 5.5 | 70.6% |
Cognition co-founder Walden Yan reported a FrontierCode score of 66.2 for Devin Fusion with Haiku 5.5 as a helper. In his assessment, the pairing reduced costs and response times.
Devin CLI supports a pairing with Opus 5.5 as the lead model and Haiku 5.5 as a helper.