Official and inference-provider prices are cash; relay prices are site credit (USD credit or CNY credit), a different unit, compared separately. Relay prices are per million total tokens at 75% input and 25% output.
Anthropic · claude-sonnet-4-5-20250929 official price: input 3, output 15 USD per million tokens (Anthropic, observed 2026-10-24). Cheapest relay: amapi.ai, group "cc" 6 USD_CREDIT per million total tokens (credit price, not cash, observed 2026-10-23). 279 sites have comparable prices; 470 sites list this model.Relay blended price uses 75% input and 25% output. Conditions: section: Model pricing · 1-hour cache write 6 USD per MTok; cacheWrite is the 5-minute write
Price ranges per million tokens
Official cash prices · Lowest 6 · median 6 · highest 6 · 1 quotes USD / per million tokens
Relay USD credit prices · Lowest 6 · median 24.6 · highest 43.2 · 2 quotes USD / per million tokens, site credit
Official price
Provider
Tariff
Input
Output
Cache read
Cache write
Unit
Observed
Conditions and source
Anthropic
Standard
3
15
0.3
3.75
USD / 1M tokens
section: Model pricing · 1-hour cache write 6 USD per MTok; cacheWrite is the 5-minute writeFull price conditions
{
"sourceUnit": null,
"sourceAmounts": null,
"conditions": [],
"contextTier": null,
"processingMode": null,
"conditionNotes": [
"section: Model pricing",
"1-hour cache write 6 USD per MTok; cacheWrite is the 5-minute write"
],
"effectiveFrom": null,
"effectiveUntil": null
}
Official prices are the vendor's cash prices. Inference providers also list cash prices and can be compared directly. Relays list prices in site credit; what you actually pay depends on each site's top-up rate, so they cannot be compared with official prices directly.
These plans are marked as test, internal, disabled, gated by top-up or membership, or have a near-zero ratio. They may not be purchasable or stable, so they are listed separately and left out of the ranking and statistics.
These are special editions of the model (thinking, max, search, batch, long context, free, etc.). They are priced differently from the base model, so they are listed separately and not ranked. Names that only add an effort level such as low, high or nothinking are compared with the base model.
All per-token prices: 93,968 prices from 854 sites (about 751 companies). 84,786 are ranked, from 838 sites (about 736 companies); 9,182 more are listed separately and not ranked. Prices collected up to 2026-10-24.
This model has 2183 recorded prices; 1022 can be compared directly (785 ranked, 237 listed separately). Not comparable yet: price not parsed yet 514, tiered by rules (see Tiered prices) 281, new price under review; showing the previous one 113, scope of the official doc price not confirmed 48, ratio looks unset 45, billed per request or another unit 42, display unit not confirmed 36
2 recent · 0 stale (older than 10 days, or not updated by this refresh) · 0 unverified timestamps. Minimum and median only count recent paid prices; older prices stay in the table. 0 zero-valued pairs are separately unverified, without proof of free eligibility. Timing does not verify payment, upstream identity or quality.
Prices are converted to site credits using each site’s published ratio. What you actually pay depends on its top-up rate. Input, output and cache prices in each row come from the same plan. Several sites run by the same company take one place in the ranking: its cheapest plan, with the other sites listed under it. We judge “same company” by identical prices, status pages or homepage code; this is not a legal finding. To see every site’s prices, choose “All plans” or filter by provider.
“Self-declared unofficial channel” means the plan name itself says “reverse-engineered” or similar. We have not verified their source or quality. Minimum and median only count regular plans.