Models

The model you choose in the session composer writes WeHarness’s replies and decides its steps in that session. Gemini 3.8 Flash is the default.

Choosing a model

• The picker lists the models WeHarness supports, grouped by provider. A model you can choose shows its name; the others show coming soon.
• Gemini 3.8 Flash and Claude Sonnet 5.5 are available now. The other models in the picker are coming soon.
• The model stays with the session until you change it. Plans are still checked by WeHarness’s decision engine, whichever model you choose.

How a model becomes available

A model is offered only when all of these hold:

1. WeHarness’s account with the provider can call it, and it answers as that model.
2. WeHarness supports it: its tool calls, structured output and streaming work with the session’s tools.
3. It passed the session agent’s full evaluation set.
4. It has a price, so its use can be counted in credits.
5. WeHarness has decided to offer it.

If one of these stops holding, for example when a provider withdraws access, the model shows as coming soon again. A session set to it then says so and does not reply, rather than moving to another model on its own.

Switching while WeHarness works

• You can pick another model while WeHarness is replying. It finishes the step it is on, then continues with the new model, and the conversation says “Switched to … for the next steps.”
• The new model sees the session as WeHarness records it: your messages, what was read, the commands that ran and their results. Nothing internal to the previous model is passed on.
• If the model you pick is not available, WeHarness says so and continues with the current one.
• WeHarness never moves your work to another model because a provider is down. It tells you, and you choose.

Credits

A message is counted at the price of the model that answered it. After a switch, each part counts at its own model’s price. A message no model answered uses no credits.

Where requests go

Each provider and where its requests are processed are listed on Service Providers and Data Locations. Code that runs in a session’s sandbox calls models through WeHarness, never with WeHarness’s keys; see Cloud work.