Claude Haiku 5.5 Joins the Fast Tier
October 8, 2026 · InfraScout Team
Claude Haiku 5.5 is now selectable on Anthropic connections. It is the fast, low-cost model for the work where the judgment is already supplied: playbook-driven runs, operator-led troubleshooting, inventory questions and quick lookups. It brings a 1M-token context window, up to 128K output tokens, and adaptive thinking.
What it costs
Haiku 5.5 is priced by the size of each request's prompt:
| Prompt up to 100K tokens | Prompt over 100K tokens | |
|---|---|---|
| Input, per million tokens | $0.10 | $0.50 |
| Output, per million tokens | $0.50 | $2.50 |
The higher rate applies to the whole request, not just the part above 100K, and InfraScout's cost estimate counts cached tokens toward the threshold. Even above the line, the per-token rates are half of Haiku 4.5's.
Because every turn re-sends the conversation so far, a long run that crosses 100K tokens pays the higher rate on every turn after that. So InfraScout keeps Haiku 5.5 conversations compacted at roughly the size a Haiku 4.5 conversation was held to, rather than letting them grow toward the full million tokens where every turn would cost five times as much. Short and focused conversations stay at the low rate. For a broad sweep across a large fleet, Choosing a Model explains when to split the run or move it to the balanced tier.
Effort works on Haiku now
Haiku 4.5 had no effort setting, so the chat's effort control was disabled on it. Haiku 5.5 supports it and starts at medium on Auto. It reasons before it answers by default, so for a lookup or an inventory question, set the chat to Low effort and keep the turn short and cheap.
Your existing connections
Haiku 5.5 replaces Haiku 4.5 in the model picker. A connection already on Haiku 4.5 keeps working and keeps costing what it always did, and the portal marks it with an Outdated model badge that names Haiku 5.5 as the replacement. Nothing moves automatically: an administrator edits the connection under AI Provider Connections and picks Haiku 5.5 when they are ready.
Questions or feedback? Reach us at info@infrascout.cloud.