[BUGF] Fix reasoning agent model compatibility and HierarchicalSwarm error propagation

[Steve-Dusty]

This PR addresses two unrelated issues:

1. Reasoning agents failing with unavailable Claude model - switches default from claude-3-5-sonnet-20240620 to
   gpt-4o
2. HierarchicalSwarm silently swallowing validation errors - ensures ValueError exceptions properly propagate

Problem 1: Reasoning Agent Model Not Found

Error

litellm.exceptions.NotFoundError: AnthropicException - {
"type": "error",
"error": {
"type": "not_found_error",
"message": "model: claude-3-5-sonnet-20240620"
}
}

Root Cause: The default reasoning_model_name parameter used claude-3-5-sonnet-20240620, which is not available or has
been deprecated by Anthropic.

Location:

• swarms/agents/reasoning_agents.py:93
• swarms/agents/reasoning_duo.py:40

Solution

Changed default reasoning model to gpt-4o, which is stable and widely available:

Before

reasoning_model_name: Optional[str] = "claude-3-5-sonnet-20240620"

After

reasoning_model_name: Optional[str] = "gpt-4o"

Rationale:

• gpt-4o is reliable and consistently available
• Other reasoning agents in the codebase already use GPT models as defaults
• Users can still override with their preferred model if needed

Problem 2: HierarchicalSwarm Exception Swallowing

Error

def test_hierarchical_swarm_error_handling():
try:
HierarchicalSwarm(agents=[])
assert False, "Should have raised ValueError for empty agents list"
except ValueError:
pass # Expected

Test FAILED - ValueError was never raised

Root Cause: The reliability_checks() method caught exceptions for logging but failed to re-raise them:

def reliability_checks(self):
try:
if not self.agents or len(self.agents) == 0:
raise ValueError("No agents found...")
except Exception as e:
logger.error(f"[ERROR] Reliability checks failed: {str(e)}")
# Missing: raise ← Exception swallowed here!

Location: swarms/structs/hiearchical_swarm.py:917

Solution

Added raise statement to propagate the exception after logging:

except Exception as e:
logger.error(f"[ERROR] Reliability checks failed: {str(e)}")
raise # ← Added this line

Impact:

• Validation errors now properly fail initialization (as designed)
• Error messages are still logged for debugging
• Tests correctly catch validation errors

Changes Made

Files Modified

File	Lines Changed	Purpose
swarms/agents/reasoning_agents.py	1	Change default model to gpt-4o
swarms/agents/reasoning_duo.py	1	Change default model to gpt-4o
test_reasoning_agent_router.py	2	Update test configs to use gpt-4o
swarms/structs/hiearchical_swarm.py	1	Add missing raise statement

Testing

Reasoning Agent Fix:

Before: NotFoundError from Anthropic

After: Successfully uses gpt-4o model

HierarchicalSwarm Fix:
$ pytest test_hierarchical_swarm.py::test_hierarchical_swarm_error_handling -v
test_hierarchical_swarm_error_handling PASSED ✅

Breaking Changes

None. Both fixes are backwards compatible:

• Users can still specify Claude models explicitly if they have access
• HierarchicalSwarm validation behavior remains unchanged (still validates at init)

Benefits

1. ✅ Reasoning agents work out-of-the-box without model access issues
2. ✅ HierarchicalSwarm validation errors are properly surfaced
3. ✅ Better error visibility for debugging
4. ✅ Tests accurately validate error handling

Checklist

• Code changes implemented
• Tests pass
• No breaking changes
• Error handling improved
• Default models use reliable options

---

📚 Documentation preview 📚: https://swarms--1213.org.readthedocs.build/en/1213/