Skip to main content
They provide you granular control over the trade-offs between quality, agentic capabilities, credit consumption and speed. There are three available variants:
VariantContext WindowCost FactorDescription
Ultra200K2xBest for hard tasks and deep reasoning. Uses Opus 4.5
High200K1x (baseline)Excels in complex problem-solving and reasoning
Medium200K0.5xBalances cost-efficiency with quality output
Mini1M0.33xHandles high volume, low-complexity tasks
You can switch between Agent Variants in further conversations in the same chat.
Each Agent Variant is powered by a continuously optimized and benchmarked model stack. We benchmark the latest LLMs from top providers across a wide range of frontend tasks - from interpretation of complex UI logic to multi-file refactors. Based on these benchmarks, Kombai’s internal model router automatically selects the ideal model for the task within the cost tier you choose. Agent Variants offer you granular control over cost. They also ensure that you consistently get the best performance per dollar, and as newer, better models emerge, without any effort to manually test and benchmark new models.

​
Extended thinking

Extended thinking helps the Agent think more deeply about your task. It is turned on by default in Kombai. You can toggle extended thinking under the Agent Variants dropdown.

​
Capabilities

  • Smarter planning: Before writing a single line, the agent thinks through implementation details, such as which components to reuse, which coding guidelines to apply, which tools to call, and which skills to fetch.
  • Better reasoning: It mentally simulates different ways to solve your problem, identifying potential edge cases and logic gaps to ensure the final solution is robust.
  • Continuous re-evaluation: The thinking doesn’t stop at the start. After every tool call, the agent pauses to verify its work against the original plan.
For Figma inputs, Kombai automatically adjusts its processing based on visual complexity. However, extended thinking adds an extra layer of architectural planning on top of that baseline. It’s best to turn on extended thinking when you are working on complex logic, major refactoring, or when architectural accuracy is crucial. Turn it off when you are working on straightforward UI tweaks where speed is the priority.
While extended thinking increases generation time and cost, it does not consume your context window, allowing the agent to think deeply without losing track of large codebases.