API429 / Pricing research

Claude API pricing: Anthropic, OpenRouter and API429

Separate base tokens from caching operations when budgeting Claude. Each comparison uses an exact model: Fable, Opus, Sonnet and Haiku are not interchangeable simply because they share a provider.

Illustration of comparing API image-generation prices
Compare the same model, workload and billing unit.

Compare prices for the same model

USD per 1 million uncached text tokens. Each cell lists input / output. Short-context Standard rates; cache, tools, regional fees, taxes and funding fees are excluded.

Exact modelOfficial providerOpenRouterAPI429
Claude Fable 5claude-fable-5$10 / $50Standard; base tokensSource$10 / $50Public catalog rateSource$3 / $15Base rate; full conditions below
Claude Fable 5.1claude-fable-5.1$10 / $50Standard; base tokensSource$10 / $50Public catalog rateSource$3 / $15Base rate; full conditions below
Claude Haiku 4 5 20251001claude-haiku-4-5-20251001$1 / $5Claude Haiku 4.5; StandardSourceNo verified exact match$0.3 / $1.5Base rate; full conditions below
Claude Opus 4 6claude-opus-4-6$5 / $25Standard; base tokensSourceNo verified exact match$1.5 / $7.5Base rate; full conditions below
Claude Opus 4 8claude-opus-4-8$5 / $25Standard; base tokensSourceNo verified exact match$1.5 / $7.5Base rate; full conditions below
Claude Opus 5claude-opus-5$5 / $25Standard; base tokensSource$5 / $25Public catalog rateSource$1.5 / $7.5Base rate; full conditions below
Claude Sonnet 4 6claude-sonnet-4-6$3 / $15Standard; base tokensSourceNo verified exact match$0.9 / $4.5Base rate; full conditions below
Claude Sonnet 5claude-sonnet-5$2 / $10Standard; base tokensSource$2 / $10Public catalog rateSource$0.6 / $3Base rate; full conditions below

OpenRouter amounts are model-list rates, excluding its separately listed platform fees. A missing quote means it was not verified for this comparison; it does not establish that the provider has no such model. Unverified reference figures are not filled with older or inferred prices.

The same workload, three budgets

Example: claude-sonnet-5, 10 million input and 2 million output tokens across many short-context requests. No cache, tools or extra operations; this is not one 10-million-token prompt.

Official provider$40
OpenRouter$40
API429$12

A changed token mix or context tier changes the result. For actual budgets, add billable reasoning, cache writes and tool use from your usage records.

Caching is a separate Claude budget item

Anthropic cache reads, five-minute writes and one-hour writes have different rates. Comparing a cheap cache hit with another service’s uncached input is misleading. The baseline workload below excludes caching; the full API429 table lists cache operations separately.

Choosing a family for the workload

Start with your own task set: coding, extraction, long documents or short classification. Compare accepted outputs, latency and cost per completed task. A rate table measures token cost rather than model quality or speed.

Tokenization can differ between model generations. The same text may produce different token counts, so inspect usage on a representative task. Direct Anthropic, OpenRouter and API429 also have separate keys, billing and feature availability.

All published Anthropic / Claude rates in API429

These rows come from the live public API429 catalog, including conditional rates. Publication is separate from access under your key. GET /v1/models confirms your permitted models.

ModelRate and condition
Claude Fable 5claude-fable-5
$3 / 1,000,000 cache-write input tokens
$3 / 1,000,000 cached input tokens
$3 / 1,000,000 input tokens
$15 / 1,000,000 output tokens
Claude Fable 5.1claude-fable-5.1
$3 / 1,000,000 cache-write input tokens
$0.075 / 1,000,000 cached input tokens
$3 / 1,000,000 input tokens
$15 / 1,000,000 output tokens
Claude Haiku 4 5 20251001claude-haiku-4-5-20251001
$0.3 / 1,000,000 cache-write input tokens
$0.3 / 1,000,000 cached input tokens
$0.3 / 1,000,000 input tokens
$1.5 / 1,000,000 output tokens
Claude Opus 4 6claude-opus-4-6
$1.5 / 1,000,000 cache-write input tokens
$1.5 / 1,000,000 cached input tokens
$1.5 / 1,000,000 input tokens
$7.5 / 1,000,000 output tokens
Claude Opus 4 8claude-opus-4-8
$1.5 / 1,000,000 cache-write input tokens
$1.5 / 1,000,000 cached input tokens
$1.5 / 1,000,000 input tokens
$7.5 / 1,000,000 output tokens
Claude Opus 5claude-opus-5
$1.5 / 1,000,000 cache-write input tokens
$1.5 / 1,000,000 cached input tokens
$1.5 / 1,000,000 input tokens
$7.5 / 1,000,000 output tokens
Claude Sonnet 4 6claude-sonnet-4-6
$0.9 / 1,000,000 cache-write input tokens
$0.9 / 1,000,000 cached input tokens
$0.9 / 1,000,000 input tokens
$4.5 / 1,000,000 output tokens
Claude Sonnet 5claude-sonnet-5
$0.6 / 1,000,000 cache-write input tokens
$0.6 / 1,000,000 cached input tokens
$0.6 / 1,000,000 input tokens
$3 / 1,000,000 output tokens

API429 catalog generated: . Public source.

Connect the model and check the result

Create an API429 key and check the permitted model IDs before migrating a workload. Keep the key on your server. Review the request contract and inspect actual usage and charges on a small representative job before increasing volume.

Sources and comparison method

Reference rates checked October 2, 2026. Recheck after a model release, promotion or pricing change. Reference figures are hidden after their review deadline; current API429 prices load from the public catalog. Neither individual contracts nor discounts are inferred.

Telegram