Claude API pricing by model and provider

Route-specific buying guide

How to budget Claude Sonnet 5 on OpenRouter

Route ID: anthropic/claude-sonnet-5

The accepted rows describe OpenRouter's Claude Sonnet 5 route. They are not Anthropic direct-list prices and cannot support a universal cheapest Claude provider claim.

OpenRouter reported a 1,000,000-token context window when checked. Long-context use, cache writes, cache reads, and tool calls need separate budget lines instead of one blended token rate.

Cost formula

Estimate uncached input, cached input, and output from the separately displayed meters. Add cache-write and web-search charges from the live route response when used; do not substitute the lower cache-read rate for cache creation.

Costs that move the bill

• Uncached context, cache-write duration, cache-hit rate, and prompt churn
• Output length, extended workflows, tool loops, retries, and fallbacks
• Separately metered web search and other enabled route features
• Router terms versus Anthropic direct terms, limits, regions, and support

Checks before purchase

• Record anthropic/claude-sonnet-5 rather than comparing only the Claude family name.
• Model cache creation and cache reads separately over the expected conversation lifetime.
• Measure cost per accepted task with real output caps, tools, retries, and failures.
• Verify the live route response before purchase because optional meters can change independently.

Source checked Aug 22, 2026: OpenRouter public model API. Recheck the live route before committing spend.

Estimate this route with the LLM workload calculator →

Compare another model family

• Qwen API pricing
• Kimi API pricing
• Grok API pricing

Decision guide

How to compare Claude API providers

Claude API is a family-level buying decision covering models, context bands, cache operations, batch service, and direct or routed access. This page attaches every price to an offered model ID. The current Claude Sonnet 5 evidence also carries an effective-date boundary, because a temporary rate must not be presented as permanent.

A simple input-output pair is only the starting point. Prompt-cache writes and reads, long context, tool use, retries, and batch eligibility can materially alter the bill. Read the provider documentation for the exact endpoint and calculate the distribution of requests your application actually sends.

What changes Claude API cost

• Exact Claude model, effective date, context length, and output ratio
• Cache write and read behavior, batch eligibility, and service priority
• Tool loops, retries, regional availability, data retention, and support

Integration checks before production

• Version the selected Claude model and any scheduled pricing change
• Apply request timeouts and bounded retries only to operations that are safe to repeat
• Capture provider request IDs and token meters without logging customer secrets

Workloads this comparison can inform

Long-document analysis

Context thresholds and cache reuse can dominate the headline input rate.

Coding assistants

Model output length, tool results, and repeated project context should be included in the estimate.

Router-based continuity

Compare the route's model identity, fallback behavior, limits, and privacy terms before treating it as direct equivalence.

How to read the price result

The direct and router observations describe Claude Sonnet 5 at the current effective rate. APIDir shows the match but withholds a cheapest claim because only two provider channels qualify.

Related model price comparisons

• GPT API pricing comparisonCompare GPT API routes by exact model ID, context band, input, cached-input and output meters, provider channel, and source.
• Gemini API pricing comparisonCompare Gemini API routes by exact model, service channel, standard or batch tier, token meter, and first-party source.
• Kimi API pricing comparisonCompare Kimi API pricing across provider routes without mixing Kimi versions, token meters, context tiers, or service channels.