From the source
From the source
From the source

From the source
You can omit max_output_tokens when using any model, including Anthropic models. Each generation can produce up to 128,000 output tokens, capped by the model's supported limit. Without an explicit budget, output tokens from earlier steps do not reduce the next generation's limit.
The new 2026-11-11.0 versions of fast , low , medium , high , xhigh , and ultra use this default. Calling an effort level by name, such as effort="fast" , selects the new version. Earlier pinned versions keep their configured budgets. Set max_output_tokens only when you want a total output-token budget across the run. See Effort and Models .
docs.perplexity.ai