Choose a GPT, GLM or DeepSeek model for your API
Evaluate capabilities, quality, latency and request cost on your own tasks.
Start with the task
Test code changes and tool calls for coding tasks, or instruction following and language quality for text tasks. Then confirm the protocol and streaming support.
Use exact model IDs
Versions and reasoning modes differ. Select an ID from the available catalog. Unaccepted models are not enabled for service.
Use comparable examples
Measure correctness, time to first response, total latency and final charge on multiple typical tasks. One successful answer does not establish reliability.