Choose the model that works best for you, or let us choose for you.

Model routing is choosing which AI model handles a task based on cost and quality, instead of sending every request to the same model. A model router, also called an LLM router, sends routine work to cheaper models and keeps frontier models for work that needs them. In Coworker, the Optimized setting does this by default, and your team can pick a different model whenever they want. To compare model prices, try the LLM cost calculator.

Coworker cuts token spend two ways. Optimized, the default, runs chats on GLM 5.3 Flash, an open-weight model priced at a fraction of Opus 5.5 and GPT 6 Astra, and people switch to a frontier model when they need one. OM2 trims context too: 9x lower token cost, and preferred 84.5% of the time in Coworker's blind review.

Coworker lets admins set a lower-cost default instead of removing choice. New chats start on Optimized, an open-weight model, and the rest of the lineup stays in the picker for the tasks that need it. Admins can set the workspace default and block specific models for everyone.

Coworker makes the lower-cost model the default: every new chat starts on Optimized, which runs on GLM 5.3 Flash, an open-weight model tuned for the lowest cost per task at frontier quality. When a task is hard, anyone can switch that chat to a frontier model such as Opus 5.5. Admins set the workspace default.

Coworker works across 13 models from Anthropic, OpenAI, Google, Moonshot and Z.ai in one picker, and people can switch models at any time. Coworker also runs inside Claude, ChatGPT and Cursor through Coworker MCP, and fails over to the next model in line if a provider is down or rate-limited.

Coworker does this with OM2, its organizational memory. Before a model sees a task, OM2 narrows it to the records that matter across your connected tools, so no model pays for tokens it doesn't use. In Coworker's 100-task benchmark against Claude's native tooling, OM2 alone cut token cost 9x and finished tasks 64% faster.

Optimized runs your work on the open-weight model Coworker rates best for the cost, tuned for the lowest cost per task completed at frontier quality. Today that's GLM 5.3 Flash from Z.ai. We re-test the pick as new models ship and update it for you, with nothing to change on your side.

An LLM gateway is the layer that connects you to many model providers through one place, with access, limits and logs managed centrally. Model routing is the choice of which model handles the work. Coworker offers both; see Coworker as an LLM gateway.

Yes. Anyone can pick any of the 13 models in the picker for a chat, such as Opus 5.5 for harder work, and switch back to Optimized at any time. Admins can also set a model on individual agents.

Optimized applies in Coworker's own apps. Through the MCP, your AI tool's own model does the reasoning and Coworker supplies the tools and context. In Coworker Code, engineers pick the model for each task.

Coworker automatically moves the request to the next model in line, whether the provider is down, overloaded or rate-limited. If a model fails partway through a response, the next model picks up where it stopped.