import type { Api, Context, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from "../types.js";
/** Approximate input token count for a request context (system + messages). */
export declare function estimateContextTokens(context: Context): number;
/**
 * Resolve the per-request output token budget.
 *
 * The model's own `maxTokens` is the ceiling. When a request context is
 * available and the model advertises a context window, we additionally cap so
 * `input + output` fits the window (a hard API requirement for Anthropic,
 * Bedrock and Gemini), leaving a small margin for estimate error and never
 * dropping below MIN_OUTPUT_TOKENS. This replaces a flat 32k clamp that both
 * throttled large-output models and ignored the window constraint entirely.
 */
export declare function resolveMaxOutputTokens(model: Model<Api>, context?: Context): number;
export declare function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string, context?: Context): StreamOptions;
export declare function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, "xhigh"> | undefined;
export declare function adjustMaxTokensForThinking(baseMaxTokens: number, modelMaxTokens: number, reasoningLevel: ThinkingLevel, customBudgets?: ThinkingBudgets): {
    maxTokens: number;
    thinkingBudget: number;
};
//# sourceMappingURL=simple-options.d.ts.map