Claude, Kimi and GLM models for AI sessions
AI sessions can now run on Anthropic's Claude, Moonshot's Kimi and Z.ai's GLM models, on the platform's keys or with a provider key your workspace connects.
Sessions used to speak only the OpenAI chat format, so a Claude key could not run one. Sessions now talk to Claude through Anthropic's native Messages API, including tool use, streaming and token usage.
Three providers are now available for sessions:
- Anthropic (Claude), with Sonnet as the default model and Opus and Haiku in the catalog.
- Moonshot (Kimi), including the Kimi K2 family.
- Z.ai (GLM), including GLM 5.3.
You can use them in two ways. Connect your own key under the workspace's model providers, and that provider bills you directly. Or run on the platform's models, where these providers sit behind the model tiers and count against your plan's allowance.
Failed or overloaded requests are retried under the same rules as before: a successful response is never sent twice, and a broken stream still records the tokens it used. See sessions for how a session runs.