Vichar Docs

Prompt caching

Cached input tokens and discounted pricing on repeated context.

Providers cache repeated prompt prefixes. When a request hits a provider-side prompt cache, the cached portion of your input tokens is billed at the model's lower cached-input rate.

  • Pricing: GET /v1/models exposes a cached input price per model where providers offer one. Cached tokens cost less than fresh input tokens.
  • Usage detail: responses report usage.prompt_tokens_details.cached_tokens, and cost_details.cached_input_cost shows what the cached share cost.
  • No code changes are required: caching is automatic where the provider supports it. Longer shared prefixes (system prompts, few-shot examples) hit more often.

Cache behavior and granularity depend on the underlying provider; treat it as a discount opportunity, not a correctness mechanism.

How is this guide?

Reasoning

Reasoning effort controls on reasoning models.

Anthropic API compatibility

Call Vichar through the /v1/messages endpoint.