Anthropic¶
from promptcanary import LiteLLMProvider
provider = LiteLLMProvider("anthropic/claude-sonnet-4-6", temperature=0.0)
Recommended Models¶
| Tier | Model string | Notes |
|---|---|---|
| Flagship | anthropic/claude-opus-4-8 |
Highest capability tier. |
| Balanced | anthropic/claude-sonnet-4-6 |
Strong default for most canary suites. |
CLI Usage¶
promptcanary run --provider anthropic/claude-sonnet-4-6 --save-baseline
promptcanary compare --provider anthropic/claude-sonnet-4-6 --fail-on-drift
A Note on Reasoning-Style Probes¶
Anthropic models historically default to more structured, step-by-step
reasoning on complex prompts than some competitors. If your suite uses
StepByStepProbe(expect_steps=False), calibrate your baseline against
Anthropic specifically — what counts as "verbose" or "direct" varies
meaningfully across providers, which is exactly why baselines are
per-provider rather than universal.