Reasoning Modes

Crucible exposes three reasoning modes that let you control the tradeoff between speed, depth, and cost per request.

Standard

The default mode. The model processes your input and returns a response without an explicit reasoning chain. Best for classification, labeling, summarization, and extraction tasks where the answer is relatively direct.

Deep

The model works through the problem step by step before producing a final output. The full reasoning chain is returned in the reasoning_trace field of the response. Deep mode is best for multi-step analysis, complex document review, and tasks where accuracy matters more than latency.

Fast

Optimized for latency-sensitive applications. The model produces a response with minimal processing overhead. Best for high-volume, lower-complexity tasks where throughput is the priority.

Choosing a mode

You can switch modes per request by passing the reasoning_mode parameter in your API call.