Models

A model is the AI system Cosine uses to understand your request, reason about the codebase, choose tools, and produce responses or code changes.

Cosine is the agentic workflow layer around the model. It provides tasks, sessions, tools, memory, environments, review flows, and integrations. The model provides the underlying intelligence used inside that workflow.

Why model choice matters

Different models have different strengths. Some are faster and cheaper, while others are better suited to long-context, ambiguous, or correctness-heavy work.

Model choice can affect:

• Speed — How quickly Cosine responds or starts using tools
• Depth — How well the model handles ambiguity, planning, and complex reasoning
• Context — How much information the model can work with at once
• Cost — How many credits the task consumes
• Reasoning support — Which reasoning levels are available for that model

Models and reasoning

You usually select a reasoning level alongside the model. The model determines the underlying AI system, while reasoning controls how much internal analysis Cosine asks that model to spend on the task.

For example:

• A faster model with low reasoning can be a good fit for simple lookups or small edits
• A stronger model with high reasoning can be a better fit for risky refactors, debugging, or design work
• Supported adaptive reasoning models can vary their thinking effort based on the task

See Reasoning for how to choose a reasoning level.

Auto model selection

Cosine can also offer an Auto model option. Auto lets Cosine treat model choice as a routing decision instead of a fixed override. This is useful when you are not sure which model is best for the task, or when a task may start simple and become more complex.

Auto is still part of the Cosine-managed model catalog. It is not a separate provider and does not require a third-party model account.

Models and cost

Models consume credits at different rates. Cosine represents this with a multiplier so models are easier to compare in the picker.

The multiplier is a pricing weight, not a quality score. A higher multiplier usually means the model is more expensive to run, not that it is always better for every task.

For exact model availability, multipliers, and billing details, see Model pricing.

Related pages

• Reasoning — How reasoning levels change speed, depth, and cost
• Model pricing — Model catalog, multipliers, and billing details
• Tasks and sessions — How model and reasoning settings fit into a session