Skip to content

The only tool you need to know what is happening and how to fix it.

Evals haiku reasoning upped

narrow-adventure49c921398af2
diff --git a/cli/test/evals/models.yaml b/cli/test/evals/models.yaml
index 3cad0357..f8be12cc 100644
--- a/cli/test/evals/models.yaml
+++ b/cli/test/evals/models.yaml
@@ -48,14 +48,15 @@ agents:
fallback_price: {prompt: 10.00, completion: 50.00, cache_read: 1.00, cache_write: 12.50}
haiku:
+ # Same settings as opus so the two compare on equal terms.
id: anthropic/claude-haiku-5.5
- label: Claude Haiku 5.5 (local smoke test)
+ label: Claude Haiku 5.5
provider:
order: [anthropic]
allow_fallbacks: true
caching: anthropic
- reasoning_effort: low
- max_tokens_per_turn: 8000
+ reasoning_effort: medium
+ max_tokens_per_turn: 16000
estimate:
input_tokens: [100000, 400000]
cached_fraction: 0.7