How Much Does Claude Sonnet Cost | API and Pro/Max Monthly Plans Explained

Quick Summary — Sonnet 4.6 Costs $3/$15 via API, from $20/month on Pro

There are two main ways to use Claude Sonnet 4.6, each with a different pricing model: pay-as-you-go via API, or subscribe to a Claude plan (Pro / Max / Team / Enterprise) and use it through the chat UI. Key points for this section are summarized below.

Source: Anthropic Pricing / Claude Docs Pricing (accessed: 2026-05-17)

The simple decision rule is: use the API if you're calling Sonnet from your own code, and a subscription if you're chatting with Sonnet through the browser or official apps. The API is easy to get started with since you only pay for what you use, but costs can grow quickly if you underestimate token usage. Subscriptions offer predictable monthly billing, though Pro and Max differ in the amount of compute available (i.e., access to Sonnet and other high-performance models).

Claude Sonnet 4.6 API Pricing — $3 Input / $15 Output and Caching Discounts

When using Claude Sonnet 4.6 via API, pricing is based on the number of tokens processed per million tokens (MTok) for both input and output. Key points for this section are summarized below.

Source: Claude Docs — Pricing (accessed: 2026-05-17)

The impact of prompt caching is particularly noteworthy. For apps that repeatedly send long system prompts or large reference documents, enabling caching means the second request onward reads the same content at just 10% of the standard rate (i.e., $0.30/MTok for Sonnet). A single cache hit with the 5-minute cache is enough to recover the write cost, and even the 1-hour cache pays for itself after two hits.

If you're new to the concept of tokens: in English, roughly "1 token ≈ 4 characters ≈ 0.75 words"; in Japanese, approximately "1 character ≈ 1–1.5 tokens." For example, sending a 10,000-character Japanese article as input would be roughly 10,000–15,000 tokens, costing around $0.03–$0.045.

Using Sonnet on a Subscription — Pro $20, Max $100, Team $25/seat

If you're using Sonnet through the Claude.ai chat UI (browser, desktop app, or mobile app), you'll need a monthly subscription plan. Key points for this section are summarized below.

Source: Anthropic Pricing (accessed: 2026-05-17)

*Max 5x and Max 20x are both listed as "From $100/month" on the official page, but are priced differently. Always check the official page for the exact monthly rate.

Here are guidelines for choosing the right plan:

1. If you just want to try Sonnet occasionally, Free is sufficient (though usage limits are strict)
2. For daily professional use, Pro ($20/month) is the minimum — Claude Code, unlimited Projects, and Research are all unlocked starting at Pro
3. If you spend hours each day conversing with Sonnet or running Claude Code for extended periods, consider Max
4. Organizations with 5+ users who want to move away from a shared account should use Team (includes SSO and audit logs)
5. Industries requiring HIPAA, ZDR (Zero Data Retention), SCIM (automated identity management), or IP restrictions should use Enterprise

Sonnet 4.6 vs. Opus 4.7 vs. Haiku 4.5 — Price Comparison

Seeing Sonnet's price in context with the higher-tier Opus and lower-tier Haiku makes it easier to assess its value. Key points for this section are summarized below.

Source: Models overview (accessed: 2026-05-17)

This table reflects information as of May 2026. Opus 4.8 has since been released, and the current top model has changed. For the latest specs and pricing, please check Models overview and claude.com/pricing.

Glossary (comparison table): Extended Thinking = a reasoning mode that outputs the thinking process explicitly / Adaptive Thinking = a feature that automatically adjusts the depth of reasoning based on the difficulty of the problem

Sonnet 4.6 is priced at 60% of Opus 4.7 (the top model at time of writing; Opus 4.8 is now the latest flagship), while sharing the same 1M token context window and supporting both Extended Thinking and Adaptive Thinking. Anthropic's official guidance positions the models as follows: Opus for pushing the frontier of complex agentic coding, Sonnet for production workloads balancing speed and intelligence, and Haiku for simple classification, summarization, and short conversations.

Previous Sonnet models (claude-sonnet-4-5 / claude-sonnet-4 / claude-3-5-sonnet) also use the same $3 input / $15 output pricing structure, meaning the Sonnet brand's price has remained consistent across generations. For more on the positioning of older models, see the related article "What Is Claude Sonnet 3.5 | Legacy Model Performance and Differences from Sonnet 4.6."

No Extra Charge for the 1M Context Window — Real Cost Examples for Long Documents

Claude Sonnet 4.6 supports a context window of up to 1M tokens (approximately 750,000 English words) at standard per-token rates — no surcharge even beyond 200K tokens. Both Opus 4.7 and Sonnet 4.6 are billed at the same per-token rate regardless of whether a request uses 900K tokens or 9K tokens, with no "long-context premium" seen in some other frontier LLMs.

Source: Long context pricing — Claude Docs (accessed: 2026-05-17)

Here is a sample cost calculation for long-document processing. If you send an entire 500,000-token input (equivalent to one full-length English novel) to Sonnet 4.6 and request a 50,000-token analysis report as output, the cost breaks down as follows:

1. Input: 500,000 tokens × $3 / MTok = $1.50
2. Output: 50,000 tokens × $15 / MTok = $0.75
3. Total: $2.25 (approximately ¥350 at $1 = ¥155)

With prompt caching enabled, if you query the same book multiple times within an hour, the input cost for the second request onward drops to $0.15 (0.30/MTok × 0.1 × 500K = $0.15). This structure makes Sonnet's pricing highly practical for analysis pipelines that repeatedly reference long documents.

50% Off with Batch API — Run Sonnet 4.6 at Half Price

For bulk processing where turnaround time is flexible, Batch API cuts Sonnet 4.6 pricing in half. Key points for this section are summarized below.

Source: Batch processing pricing — Claude Docs (accessed: 2026-05-17)

Batch API is designed for asynchronous processing with up to 24-hour turnaround, making it well-suited for scenarios where real-time responses aren't required — overnight batch jobs, large-scale document analysis, dataset generation, content rewriting, and similar tasks. The Batch discount can be stacked with the prompt caching discount, so combining "Batch + cache read" can compress input costs down to $0.15/MTok.

Additionally, the 64K token max output limit is relaxed in Batch API — by adding the output-300k-2026-03-24 beta header, Sonnet 4.6 can generate outputs up to 300K tokens long, making it well worth leveraging Sonnet for long-form report generation.

3 Ways to Reduce Sonnet Costs

Here are concrete strategies for getting the most out of Sonnet 4.6 while keeping costs under control.

1. Always Enable Prompt Caching

If you're repeatedly sending the same system prompt or reference material, simply adding cache_control drops the input rate to 10% of standard ($0.30/MTok for Sonnet). This is highly effective for stock-type use cases like chatbots, code review bots, and FAQ responders — the initial write cost (1.25x) is quickly recovered after just a few hits.

2. Route Non-Urgent Processing to Batch API

Any job that can tolerate up to 24-hour turnaround — overnight reports, dataset generation, reprocessing existing content — should be sent to Batch API, which halves both input and output costs. Stacking Batch with caching can bring Sonnet 4.6's effective cost down to a fraction of the standard rate.

3. Route Tasks to the Right Model

There's no need to run everything through Sonnet. Simple classification, short summaries, and known FAQ responses can be handled adequately by Haiku 4.5 ($1 input / $5 output) at one-third the price of Sonnet. Conversely, routing only long-running agentic coding tasks or complex research to Opus 4.7 allows you to control overall spend while maintaining quality where it matters.

Summary — Sonnet 4.6 Pricing at a Glance

In one line: Claude Sonnet 4.6 costs $3 input / $15 output per million tokens via API, and starts at Pro $20/month for subscriptions. It's 40% cheaper than Opus 4.7 and handles up to 1M tokens of long context with no additional surcharge — which is why Sonnet is easy to position as the workhorse for production workloads.

For subscription users chatting through the UI, Pro ($20/month, equivalent to $17/month annually) is the individual minimum. For API users, combining prompt caching and Batch API can significantly reduce effective costs. Since pricing is subject to change, always check the latest rates at claude.com/pricing and Claude Docs Pricing before committing to a purchase decision.

Sources and Reference Links

1. Anthropic Pricing (official pricing page) — Monthly rates for Pro / Max / Team / Enterprise
2. Claude Docs — Pricing — Detailed API, Batch, prompt caching, and 1M context pricing
3. Models overview — Spec comparison for Sonnet 4.6 / Opus 4.7 / Haiku 4.5
4. Related article: Complete Claude Pricing Comparison | How to Choose Free, Pro, or Max
5. Related article: What Is Claude Sonnet 3.5 | Legacy Model Performance and Differences from Sonnet 4.6