Claude Code · AI agents · subagents · Haiku 5.5
Background subagents in Claude Code now use less of your usage limit
A running background subagent refreshes its status line every 30 seconds. Each refresh used to re-read the subagent's whole conversation; now Haiku 5.5 writes it from short notes, about 99% cheaper per update. What changed, what the changelog says, and how to check subagent usage.

Background subagents in Claude Code now use less of your usage limit. The @ClaudeDevs account announced the change on X on 4 November 2026. A running subagent updates its status line every 30 seconds, and each update used to re-read the subagent's whole conversation. Now Haiku 5.5 writes the status line from short notes, which Anthropic says cuts the cost of each update by about 99%.
What a background subagent is
A subagent is a separate Claude worker inside your session. It gets its own context, does a side task, and returns a result to the main conversation. A foreground subagent blocks the main conversation until it finishes. A background subagent runs while you keep working. According to the subagents docs, with nothing configured, a subagent that Claude spawns from your main conversation starts in the background.
Running subagents appear in a panel below the prompt input, one row each, and /tasks lists everything running in the background of the session. The docs also say that a subagent sends its own requests, and those count toward the same usage limits as your main conversation.
Where the cost came from
While a subagent works, its row shows a short line about what it is doing, and that line refreshes every 30 seconds. To write it, Claude Code used to send the subagent's whole conversation to the model again. A subagent that has read a dozen files and run a test suite carries a long transcript, so every refresh re-read all of it, twice a minute, for as long as the subagent ran, and separately for each background subagent you had open. All of that spend went into a few words on screen.
This cost had been worked on before. The Claude Code changelog for 2.1.128 lists two fixes: progress summaries that missed the prompt cache (about a 3x reduction in cache creation, per the entry), and summaries that kept firing while a subagent's transcript wasn't changing. Version 2.1.274 fixed a progress summary being replaced by a runaway multi-paragraph reply. Those fixes made each re-read cheaper or less frequent. The new change stops re-reading the conversation at all.
What changed
Haiku 5.5 now writes the status line from short notes instead of from the conversation. Each update reads a few notes instead of the full transcript, and the model reading them is Haiku, which the Claude Code docs recommend for lower-cost work. The announcement puts the result at about 99% less per update.
Haiku 5.5 (claude-haiku-5-5) arrived in Claude Code 2.1.293 as the default Haiku model on the Anthropic API, priced at $0.10 per million input tokens and $0.50 per million output tokens.
Do you need to do anything
The announcement doesn't name a setting or a version to install. The Claude Code changelog has no matching line either: as of 2.1.318, released on 4 November 2026, no entry mentions Haiku writing subagent status lines.
The 2.1.318 release does touch the same feature in two places. It changes progress summaries on Bedrock, Vertex, Foundry, or with telemetry off, so that on a cold prompt cache one can wait up to 5 minutes for the subagent's pending response. It also fixes a "Now using usage credits" notice that reappeared about every 30 seconds while a background subagent was running. The announcement doesn't say whether the Haiku change covers the cloud providers or sessions with telemetry off, so if you run Claude Code that way, check your own numbers before counting on the saving.
How to see what subagents cost you
On a Pro, Max, Ultra, Team, or Enterprise plan, /usage shows a breakdown of recent usage attributed to skills, subagents, plugins, and individual MCP servers, each as a share of the total. Press d or w to switch between the last 24 hours and the last 7 days. The costs docs list subagents among the reasons usage climbs in a long session, because every subagent sends its own requests on top of the main conversation's.
If you build on the Agent SDK, these summaries are opt-in. The TypeScript SDK's agentProgressSummaries option defaults to false. Set to true, it generates one-line progress summaries for foreground and background subagents and forwards them on task_progress events in the summary field.
The status line was one part of a background subagent's cost. The rest is the subagent's own work, and the docs give two ways to cut it: set model: haiku in a subagent's definition for simple tasks, or set CLAUDE_CODE_SUBAGENT_MODEL to run every subagent on one model.
Sources
- @ClaudeDevs, announcement on X, 4 November 2026
- Claude Code changelog, versions 2.1.128, 2.1.274, 2.1.293 and 2.1.318
- Create custom subagents, Claude Code docs
- Manage costs effectively, Claude Code docs
- Agent SDK reference, TypeScript, Claude Code docs