Learn about this first
Claude Code's model and effort settings both aim to improve output, but they work differently. A larger model has greater capabilities, as measured by industry-standard benchmarks. Effort controls the total work Claude puts into a request, including thinking time, the number of files read, how thoroughly it verifies its work, and how deeply it pursues multi-step tasks. At high effort, Claude performs more actions (reading files, running tests, and checking again); at low effort, it is more likely to ask for context. Choosing a model essentially switches between different sets of frozen weights: the weights are fixed during training, and prompts and context can only steer the model, not change its weights. Model hallucinations are plausible but incorrect token sequences produced by the weights.
Latest developments
Claude Code's Model and Effort: Knowing More vs. Trying Harder
Event progress1
Anthropic Releases Claude Code
Claude Devs: Claude Code's Model and Effort: Knowing More vs. Trying Harder