Models

Every Claud model is reached through the same endpoint; switching models is a one-word change. This page lists the current catalogue. Prices and limits below are read live from the published catalogue and match what GET /v1/models returns.

Catalogue#

Prices are per one million tokens in USD and are converted to Claud tokens at the account rate; see Tokens & billing. Cached input is the discounted rate for prompt tokens served from the prompt cache.

Choosing a model#

• claud-5.1: the default. Strong reasoning with quick responses. Start here for chat, coding assistants, analysis and agents.
• claud-opus: our most capable model for complex, multi-step work where quality matters most: long documents, architecture decisions, difficult synthesis. Available on paid plans.
• claud-reason: extended thinking for hard problems in mathematics, research and intricate debugging. Slower and produces more reasoning tokens; best when correctness beats latency.
• claud-flash: lowest latency and cost for high-volume tasks: classification, extraction, summarisation, routing and simple chat.

Aliases#

Some models have aliases that always point at the latest version:

Responses always report the canonical slug in model, so you can see what actually served the request.

Capabilities#

• Reasoning models think before answering. Their thinking is returned in reasoning_content when present and counted in completion_tokens_details.reasoning_tokens. Control effort with reasoning_effort; see Tool calling & JSON.
• Tool calling lets the model request calls to functions you define.
• JSON mode (response_format: {"type": "json_object"}) guarantees syntactically valid JSON output.
• Vision models accept image_url content parts (data URLs or public URLs).

Context windows and output limits#

The context window is the total of prompt and completion tokens a model can handle in one request. Your plan may cap it lower: pay-as-you-go and Starter accounts are limited to 128K tokens of context regardless of model; Pro and above get the full window. Requests above the effective limit fail before billing with context_length_exceeded.

Max output is the ceiling for max_tokens. If you omit max_tokens, Claud applies a sensible per-model default (8K for claud-5.1, 16K for claud-opus and claud-reason, 4K for claud-flash) and never exceeds what fits in the remaining context.

Fallbacks#

For resilience, some models define a fallback of equal or greater capability. If the primary route is unavailable, rate-limited upstream, or times out before producing output, Claud transparently retries on the fallback. When this happens:

• model in the response still shows the model you asked for;
• streaming responses include "fallback_used": true in the final fable billing event;
• you are billed at the price of the model you requested, never more.

Deprecations#

When a model is scheduled for retirement it is marked deprecated: true in GET /v1/models and on this page, and continues to work until the announced date. Aliases are updated to point at the successor. We announce deprecations by email to all account owners at least 30 days in advance.

Listing models programmatically#

curl https://api.claudkey.com/v1/models \
  -H "Authorization: Bearer $CLAUD_API_KEY"

See the Models reference for the full response shape.