I have used Claude and GPT to ship production AI agents. At Avoca, we use Claude for our conversational agents. At Fegmo, we used Claude on AWS Bedrock. At Fuse AI, we used GPT-4.
Honest comparisons for products that are being shipped in production.
Tool calling: Claude is better
Claude is more reliable and graceful in tool calling (versus function calling for GPT). Claude almost never hallucinates tool parameters. GPT sometimes hallucinates parameters that are not in the schema.