pub const EXPECTED_RESPONSE_TOKENS: u64 = 16_384;Expand description
Visible output (text and tool calls) one round is expected to need on
top of its thinking. The output reserve is this plus the thinking reserve,
capped at the request’s max_tokens (ContextBudgetInputs::max_output_tokens).
Not the max_tokens ceiling itself (128k for Opus): Claude 4.5 and later
accept a request whose input plus max_tokens exceeds the window and
stop at the window with stop_reason: "model_context_window_exceeded"
(https://platform.claude.com/docs/en/build-with-claude/context-windows#context-window-overflow-behavior),
and for the earlier models that reject it the request builder clamps
max_tokens to the room left (HelperMethods::fit_max_tokens_to_window).
Reserving the full ceiling put hard capacity at ~31.5k on a 200k Opus
window. 16k covers a large single-file write in one round; a round’s
output becomes the next round’s input, so the following assessment sees
it either way.