vinzenz/hyperhive
vinzenz/hyperhive
Watch
0
0
Fork
You've already forked hyperhive
 0
Code Activity

subagent: haiku 5-5 is near free up to high effort on a small context, but max costs about what sonnet does

Browse source
This commit is contained in:
damocles 2026-11-01 12:24:40 +01:00 • committed by mara
commit a1f1f04657
2 changed files with 14 additions and 15 deletions
Download patch file Download diff file Expand all files Collapse all files

22
claude-plugins/plugins/base/skills/claude-subagents/SKILL.md
View file

@ -59,15 +59,13 @@ more readily than you might for your own work — a subagent is throwaway
compute, not a peer you're rationing. Haiku in particular is cheap enough compute, not a peer you're rationing. Haiku in particular is cheap enough
that you don't need to be precious about spinning one up: for a mechanical, that you don't need to be precious about spinning one up: for a mechanical,
well-scoped batch (the shape this skill is for), reach for it liberally well-scoped batch (the shape this skill is for), reach for it liberally
rather than defaulting to a bigger model "to be safe." Haiku also stays rather than defaulting to a bigger model "to be safe." On a small context
cheap regardless of `effort` level — unlike a frontier model, cost on Haiku costs next to nothing up to `high` effort, so there's no cost reason
Haiku doesn't scale with effort because effort isn't a real lever on it to cap its effort at `medium` by reflex. Effort is a real lever on Haiku
at all (measured: `thinking_tokens` doesn't visibly move between `low` though: at `max` it behaves like a slightly weaker Sonnet at about Sonnet's
and `max`). So there's no cost reason to cap Haiku's effort the way the cost, so `max` on Haiku buys nothing over picking Sonnet. Stay at `high` or
Effort section below recommends for frontier models — but don't expect below, and move to a stronger model when the batch needs more judgement
raising it to buy extra reasoning either; a Haiku batch that genuinely (see Pitfalls below).
needs more judgement wants a stronger model instead (see Pitfalls below),
not a higher effort level on the same model.
This isn't "always pick the smallest model regardless of task" — a batch This isn't "always pick the smallest model regardless of task" — a batch
that actually needs judgement calls still wants a stronger model (see that actually needs judgement calls still wants a stronger model (see
@ -80,9 +78,9 @@ An omitted `effort` defaults to `medium` here — cheaper than claude's own
model default (`high` on most models), a deliberate cost-conscious choice model default (`high` on most models), a deliberate cost-conscious choice
for subagent work specifically, same spirit as the model-choice guidance for subagent work specifically, same spirit as the model-choice guidance
above. Raise it explicitly when the task is complex enough to actually above. Raise it explicitly when the task is complex enough to actually
need deeper reasoning, not as a reflex. **Exception: on Haiku, cost doesn't need deeper reasoning, not as a reflex. **Exception: on Haiku with a small
scale with effort the way it does on a frontier model** — see Model choice context, cost stays near zero up to `high`.** See Model choice above: skip this
above, don't apply this same-default-medium caution to a Haiku subagent. default-medium caution there, but keep `max` for a stronger model.
Anthropic's own levels (per current Claude Code docs — names/availability Anthropic's own levels (per current Claude Code docs — names/availability
are model-dependent, check before relying on an exact list): `low`, are model-dependent, check before relying on an exact list): `low`,

7
hive-subagent-mcp/src/mcp.rs
View file

@ -57,9 +57,10 @@ struct StartArgs {
/// models). Independent of `model`, so a cheap model at high effort or /// models). Independent of `model`, so a cheap model at high effort or
/// an expensive one at low effort are both valid combinations, not /// an expensive one at low effort are both valid combinations, not
/// just the two extremes. See the `base:claude-subagents` skill for /// just the two extremes. See the `base:claude-subagents` skill for
/// Anthropic's own guidance on choosing between levels — note that on /// Anthropic's own guidance on choosing between levels. On Haiku with a
/// Haiku specifically, cost doesn't scale with effort the way it does /// small context, effort up to `high` costs next to nothing, so there's
/// on a frontier model, so there's no cost reason to cap it at `medium`. /// no cost reason to cap it at `medium`; at `max` it costs about what
/// Sonnet does.
#[serde(default)] #[serde(default)]
effort: Option<String>, effort: Option<String>,
/// Path to a file holding the subagent's actual task instructions. A /// Path to a file holding the subagent's actual task instructions. A