claude-sonnet-4-6 API — per-token pricing and code examples

Checking statusPrice version 3 · in effect since Nov 19, 2026

Price per 1M tokens

InputOfficial price $3.00$0.60
Cache readOfficial price $0.30$0.06
Cache writeOfficial price $3.75$0.75
Cache write 1 hOfficial price $6.00$1.20
OutputOfficial price $15.00$3.00

Minimum per request$0.00002

Cache rates apply when the provider reports cache tokens.

80% off

About this model

claude-sonnet-4-6 is served through OpenAI Chat Completions, OpenAI Responses, Anthropic Messages. You pay $0.60 per 1M input tokens and $3.00 per 1M output tokens. That is 80% below the official price of $3.00 and $15.00. Cache reads cost $0.06 per 1M tokens when the provider reports cached tokens. It is included in budget packs from $9.99 up.

Where Claude Sonnet 4.6 fits

Claude Sonnet 4.6 is the earlier Sonnet generation from Anthropic, now in legacy status. Anthropic released it on February 17, 2026 and lists its retirement as not sooner than February 17, 2027, with knowledge that runs to August 2025. It takes text and image input, replies in text, holds up to 1M tokens of context and can write up to 128K tokens in a single synchronous reply.

Anthropic's model page gives no new use-case pitch for Sonnet 4.6 and recommends Claude Sonnet 5.5 for anyone starting fresh. Sonnet 4.6 still makes sense for an app that was built and tested on it, for a gradual migration where you compare old and new answers side by side, or for traffic where you want thinking to stay off unless a request asks for it.

Set the effort yourself

Like Opus 4.6, Sonnet 4.6 does not think by default. A request with no thinking field gets a direct answer. To let the model reason, send thinking: {type: "adaptive"}. The older enabled mode with budget_tokens is still accepted but deprecated, and between_tools is rejected with 400. When thinking is on, the default display is summarized.

The part that surprises people is effort. Sonnet 4.6 defaults to high, and Anthropic's effort guide warns about it directly: "Explicitly set effort when using Sonnet 4.6 to avoid unexpected latency." The same guide describes medium as the recommended default, with the best balance of speed, cost and performance for most applications. Accepted values are low, medium, high and max; xhigh does not exist on this model.

In practice, put output_config.effort in every Sonnet 4.6 request instead of relying on the default. Use low for routing, tagging and short rewrites, medium for everyday chat and coding help, and keep high or max for the requests where a wrong answer costs more than waiting.

What it costs on Cent AI

Sonnet 4.6 draws from your prepaid USD balance per token, at the rates listed in this page's price box. When adaptive thinking is on, every reasoning token is charged at the output rate. That is why the effort advice above is also cost advice: leaving Sonnet 4.6 at its high default can make a short answer cost far more output than its visible length.

Prompt caching on Sonnet 4.6 starts at 1,024 tokens. Anthropic offers a 5-minute default lifetime and an optional 1-hour lifetime, and on Cent AI cache writes for both lifetimes are billed at the input rate. Both cache rates apply only to the cached token counts the model reports back. Changing the thinking setting or the effort level between turns invalidates cached message blocks, so pick one effort per conversation.

Using it from your tools

Sonnet 4.6 answers on /v1/messages, /v1/chat/completions and /v1/responses. Cent AI does not change your parameters silently, so the model's validation applies. Prefilling the last assistant message ends in a 400 error. Starting with API version 2026-10-21, sending any sampling value such as temperature, top_p or top_k also ends in a 400; older API versions only limited those values while thinking was on. Tool calling and structured outputs are supported, and forcing a specific tool works when you use adaptive thinking.

A /v1/messages request with effort pinned to medium:

{
  "model": "claude-sonnet-4-6",
  "max_tokens": 2048,
  "output_config": { "effort": "medium" },
  "messages": [{ "role": "user", "content": "Tag this support ticket" }]
}

Example: the cost of one request

Take a request that sends 2,000 new input tokens, reads 8,000 tokens from cache and gets 800 output tokens back. At the rates in the price box on this page:

  • input: 2,000 × $0.45 / 1,000,000 = $0.0009
  • cache read: 8,000 × $0.045 / 1,000,000 = $0.00036
  • output: 800 × $2.25 / 1,000,000 = $0.0018

That request costs $0.00306, about Rp56 at Cent AI's exchange rate on 2026-11-14 (Rp18,231.96 per USD). A thousand requests like it cost $3.06. Your real requests will differ; each one shows its own amount in usage.cost and on the Requests page.

Questions

Is there a 5-hour or weekly limit? No. You pay per token from your balance. Each key accepts up to 600 requests per minute, and you can set your own daily or monthly limit on a key.

Is a failed request charged? A request that fails before the model receives it costs $0. Tokens the model already produced are charged.

Does unused balance expire? No, not while your account exists. Every top-up is manual, with no auto-renewal.

Compare with claude-sonnet-5.5, claude-opus-4-6, claude-opus-5, or see every model on the pricing page.

Specification

Context window
—
Input and output
Text
Endpoints
/v1/chat/completions, /v1/responses, /v1/messages
Smallest pack
$9.99

Latency and error rate per model are on the status page. status.cent-ai.dev

What a request costs

Open the full calculator

Cost per request

90% is an assumption you can change, not a promise. Cache is billed only from cache tokens the model provider reports.

See how many such requests it covers.

Estimate

claude-sonnet-4-6
Fresh input99,000 tok
Cache read891,000 tok
Output10,000 tok

Per request$0.1428
At official pricesAt official prices $0.7143

Total$0.1428

An estimate at price version 3. Your bill comes from the usage the provider reports for each request.

Save $0.5714

Code

curl https://api.cent-ai.dev/v1/chat/completions \
  -H "Authorization: Bearer $CENT_AI_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-sonnet-4-6", "messages": [{"role": "user", "content": "Hello"}]}'

Set CENT_AI_KEY to a key from the dashboard. OpenAI SDKs use https://api.cent-ai.dev/v1; Anthropic SDKs use https://api.cent-ai.dev.

Price history

Version 3, in effect since Nov 19, 2026.

  1. Effective

    per 1M tokens

    Inputold $0.45 new $0.60
    Cache readold $0.045 new $0.06
    Cache writeold $0.45 new $0.75
    Cache write 1hold $0.45 new $1.20
    Outputold $2.25 new $3.00
  2. Effective

    per 1M tokens

    Inputold $0.45 new $0.45
    Cache readold $0.045 new $0.045
    Cache writeold $0.5625 new $0.45
    Cache write 1hold $0.90 new $0.45
    Outputold $2.25 new $2.25

Compare with

All models against official prices

Questions

How much does claude-sonnet-4-6 cost on Cent AI?

$0.60 per 1M input tokens and $3.00 per 1M output tokens. Cache reads cost $0.06 and cache writes $0.75 per 1M tokens. Each request is rounded up once.

Is there a minimum charge?

A successful request costs at least $0.00002. A request that ends early, because the client left or the output was cut off, pays only for its tokens.

How do I pay?

Top up a prepaid USD balance with QRIS or USDT/USDC, or buy a budget pack once. Nothing renews automatically.

What if the provider does not report usage?

The request is billed from an estimate and marked estimated on the Requests page. An estimate never contains cache tokens.

Which endpoint do I call?

/v1/chat/completions, /v1/responses, /v1/messages. The OpenAI base URL is https://api.cent-ai.dev/v1.

Try claude-sonnet-4-6 with one key

One prepaid balance for every model. Pay with QRIS or USDT/USDC.

Start free