Add Claude 4 Sonnet to evals (#68)

* Add Claude 4 Sonnet to evals * Avoid the Braintrust proxy

Author: Nicolapps

.github/workflows/daily_anthropic_evals.yml

@@ -46,7 +46,7 @@ jobs:
 
  - name: Set Anthropic models
  run: |
- echo "MODELS=claude-3-5-sonnet-latest,claude-3-7-sonnet-latest" >> $GITHUB_ENV
+ echo "MODELS=claude-3-5-sonnet-latest,claude-3-7-sonnet-latest,claude-sonnet-4-0" >> $GITHUB_ENV
 
  - name: Run evaluations
  env:

runner/eval_convex_coding.py

@@ -207,6 +207,7 @@ def convex_coding_task(model: ModelTemplate, input: str):
 model_names = [
  "claude-3-5-sonnet-latest",
  "claude-3-7-sonnet-latest",
+ "claude-sonnet-4-0",
  "gpt-4o",
  "o3-mini",
  "gemini-2.0-flash-lite",

runner/models/__init__.py

@@ -38,6 +38,15 @@ class ModelTemplate(BaseModel):
  uses_system_prompt=True,
  provider=ModelProvider.ANTHROPIC,
  ),
+ ModelTemplate(
+ name="claude-sonnet-4-0",
+ formatted_name="Claude 4 Sonnet",
+ max_concurrency=int(os.getenv("ANTHROPIC_CONCURRENCY", "2")),
+ requires_chain_of_thought=True,
+ uses_system_prompt=True,
+ provider=ModelProvider.ANTHROPIC,
+ override_proxy="https://api.anthropic.com/v1",
+ ),
  ModelTemplate(
  name="gpt-4o",
  formatted_name="GPT-4o",