Token Limit
Domain: Models & Providers
Motivation
Token limits are the bounds on how many tokens a request may carry, both for the input context and the generated output. The DeepSeek adapter declares a defaultContextWindow (1,000,000) as a capacity fallback and a maxTokens output cap (256,000); a model’s own cap and an explicit request value win over the default.
Concrete Example
A request whose retained images exceed maxRequestFilesBytes fails with IMAGE_OFFLOAD_REQUIRED; exceeding the context window fails with CONTEXT_WINDOW_EXCEEDED.
Analogy
The speed limit sign — the road sets the maximum, and a faster posted limit for this stretch wins.