400 litellm.BadRequestError: AnthropicException

[beaugast]

Session ID: v2:77d21987-0e51-4fdf-9a3b-460d5eecf7ed
Session Schema: v2.0
Pro User ID: ****iLCm

Issue Description (required)

Sorry, there was an error from the AI: [Request ID: b88f24d4-ee63-4fd8-bb80-5e4dfa8c3544] {"error":"400 litellm.BadRequestError: AnthropicException - b'{"type":"error","error":{"type":"invalid_request_error","message":"This model does not support assistant message prefill. The conversation must end with a user message."},"request_id":"req_011CZzQFLniAWVqcU9iX8vqX"}'. Received Model Group=anthropic/claude-sonnet-4-6\nAvailable Model Group Fallbacks=['anthropic/claude-sonnet-4-6:backup']\nError doing the fallback: litellm.BadRequestError: BedrockException - {"message":"The model returned the follo"}

Can we get rid of LiteLLM and use Amazon? Nothing but problems all the time with LiteLLM including recent security breaches.

Expected Behavior (required)

To work.

Actual Behavior (required)

Sorry, there was an error from the AI: [Request ID: b88f24d4-ee63-4fd8-bb80-5e4dfa8c3544] {"error":"400 litellm.BadRequestError: AnthropicException - b'{"type":"error","error":{"type":"invalid_request_error","message":"This model does not support assistant message prefill. The conversation must end with a user message."},"request_id":"req_011CZzQFLniAWVqcU9iX8vqX"}'. Received Model Group=anthropic/claude-sonnet-4-6\nAvailable Model Group Fallbacks=['anthropic/claude-sonnet-4-6:backup']\nError doing the fallback: litellm.BadRequestError: BedrockException - {"message":"The model returned the follo"}

System Information

• Dyad Version: 0.43.0
• Platform: win32
• Architecture: x64
• Node Version: v22.14.0
• PNPM Version: 10.18.3
• Node Path: C:\Program Files\nodejs\node.exe
• Pro User ID: ****iLCm
• Telemetry ID: b05f6e4b-e2b6-44d2-aaa1-151815ce1fcb
• Model: anthropic:claude-sonnet-4-6 | customId: undefined

Settings

• Selected Model: anthropic:claude-sonnet-4-6
• Chat Mode: local-agent
• Auto Approve Changes: n/a
• Dyad Pro Enabled: true
• Thinking Budget: n/a
• Runtime Mode: host
• Release Channel: beta
• Auto Fix Problems: false
• Native Git: true

Logs

  hasProviderModels: true
}
  source: 'remote',
  version: '2026-04-12T18:03:26.656Z',
  expiresAt: '2026-04-12T20:03:28.156Z'
}
  source: 'remote',
  version: '2026-04-12T18:03:26.656Z',
  providerCount: 6
}
  source: 'remote',
  version: '2026-04-12T18:03:26.656Z',
  expiresAt: '2026-04-12T20:03:28.156Z'
}
  providerId: 'anthropic',
  source: 'remote',
  version: '2026-04-12T18:03:26.656Z',
  hasProviderModels: true
}

[beaugast]

Sorry, there was an error from the AI: [Request ID: 3e1e347d-14d9-4bd3-8715-ca4f4562eca9] {"error":"400 litellm.BadRequestError: AnthropicException - b'{"type":"error","error":{"type":"invalid_request_error","message":"This model does not support assistant message prefill. The conversation must end with a user message."},"request_id":"req_011CZzQumAixhwK7qPo8HhAu"}'. Received Model Group=anthropic/claude-sonnet-4-6\nAvailable Model Group Fallbacks=['anthropic/claude-sonnet-4-6:backup']\nError doing the fallback: litellm.BadRequestError: BedrockException - {"message":"The model returned the follo"}

[wwwillchen]

Hi @beaugast - thanks for reporting this bug. Sorry about that, we've fixed a bug on the server and have deployed it. Please let us know if you see this error any more.

As a bit of context: all of our errors are wrapped with LiteLLM because we use it to route to various AI providers like Anthropic, Amazon, OpenAI, etc. (this is what allows us to support a wide range of models), so this error wasn't actually caused by LiteLLM because they are just the wrapper. In case you're wondering, we weren't affected by the recent LiteLLM security incident because we use their docker image and we also pin to a specific stable version to avoid accidental upgrades (and potentially malicious updates).

Let me know if you have any more questions and thanks for the detailed bug reports.