Quick category search Chatbots Image Generation Video Generation Audio & Voice Coding Writing Productivity Research AI Agents Free Tier Table

Ask CatAI Tool SummaryClaude

Claude Sonnet 5 Price Hike Cancelled: US$2/US$10 per MTok Is Now Permanent (2026)

文章最後更新:2026-08-16

If your product runs on the Claude Sonnet 5 API, this changes your September bill: Anthropic has cancelled the 50% price increase that was scheduled for September 1, 2026. The launch pricing of US$2 / US$10 per MTok (per million input / output tokens) is now the permanent standard price.

This is not a rumour. Here is the verbatim text from Anthropic’s official developer pricing docs (last checked 2026-08-16):

The US$2/US$10 per million input/output token pricing for Claude Sonnet 5, announced at launch as introductory pricing through August 31, 2026, is now the standard price. The previously scheduled increase to US$3/US$15 per million input/output tokens on September 1, 2026 will not occur.

In plain terms: US$2/US$10 was originally introductory pricing that expired August 31, with a jump to US$3/US$15 on September 1. Anthropic has now stated that increase will not happen.

Current official Sonnet 5 rates, column by column

Figures taken directly from Anthropic’s official pricing page and developer docs — both sources agree (checked 2026-08-16). MTok = per million tokens, USD.

ModelBase input5m cache write1h cache writeCache hits & refreshesOutput
Claude Sonnet 5US$2US$2.50US$4US$0.20US$10
Claude Sonnet 4.6US$3US$3.75US$6US$0.30US$15
Claude Haiku 4.5US$1US$1.25US$2US$0.10US$5
Claude Opus 5US$5US$6.25US$10US$0.50US$25

The Batch API takes another 50% off both input and output (official wording). That puts Sonnet 5 at US$1 / US$5 per MTok, Haiku 4.5 at US$0.50 / US$2.50, and Opus 5 at US$2.50 / US$12.50.

What the cancellation is actually worth

Straight multiplication against the official rates (this is list-price math, not a measured invoice). Assume 100M input tokens plus 10M output tokens in a month:

ScenarioInput costOutput costMonthly total
Permanent price, US$2/US$10US$200US$100US$300
Scheduled Sept 1 price, US$3/US$15US$300US$150US$450
Permanent price via Batch APIUS$100US$50US$150

Same workload, US$150 less per month than expected — a third off. The gap scales linearly with volume, which is why this matters most for agentic workloads: agents make many tool calls and re-read context repeatedly, so token consumption grows fast with task complexity and a one-third unit-price difference gets multiplied many times over.

The variable people miss: token counts differ across models

This is where the math often goes wrong, so here is the official text:

Claude 4.7 and later models and Claude Mythos Preview use a newer tokenizer that contributes to their improved performance on a wide range of tasks. This tokenizer produces approximately 30% more tokens for the same text. The exact increase depends on the content and workload shape. Claude Sonnet 4.6 and earlier models use the previous tokenizer.

Models from 4.7 onward (Sonnet 5 included) use a newer tokenizer that splits the same text into roughly 30% more tokens, with the exact figure depending on content. So there are two distinct cases:

  • Already on Sonnet 5 → the cancellation is a genuine one-third saving. Same model, same tokenizer: the list price drops, and your bill drops with it.
  • Migrating from Sonnet 4.6 to Sonnet 5 → the list price falls from US$3 to US$2 (down 33%), but the same text yields more tokens. “One-third cheaper per token” does not mean “one-third cheaper on the invoice.” You need to measure it against your own workload. Anthropic says approximately 30% and explicitly notes it varies by content, so we will not estimate it for you.

This has nothing to do with Claude subscriptions

To be clear: everything above is API usage-based pricing (developers paying per token). The Claude Pro, Max and Team monthly subscription plans are unaffected by this change — those are priced on an entirely different basis. For subscription prices, free-tier limits and known payment details, see our Claude tool page.

Sources and verification status

  • Price figures: taken from the Anthropic official pricing page and the Anthropic developer pricing docs. Both official sources agree; read directly on 2026-08-16.
  • The cancellation itself: confirmed in writing by Anthropic’s official developer docs (quoted above). Chinese-language coverage of the news appeared in this cnyes report, 2026-08-11.
  • Not verified: that same report also mentions market rumours about a next-generation Sonnet 5.5 (internal codename “Fennec”) and its context window. That is second-hand market chatter, not an official announcement. We are relaying it only, it does not represent this site’s position, and Anthropic’s official announcements govern. We could not find it on any official page, so none of its figures appear anywhere in this article.

All prices verified on 2026-08-16. API pricing can change at any time — always confirm against the official Anthropic pricing page.

How Amo and Pimi look

AMO Amo Finding faults
免費版最讓人抓狂的是額度規則——它是滾動式 5 小時窗,不是每天歸零,用完 10-20 則就得乾等,而且免費版根本不含 Claude Code。付費方案更誇張,Pro、Max 5x、Max 20x、Team Standard、Team Premium、Enterprise 六檔疊在一起,Team Premium 年繳還要 US$100/席/月。最扯的是 Max 20x 官網頁面只寫「Starting at US$100」,跟 5x 顯示一模一樣,我們查了十幾次都沒辦法從頁面文字確認實際價差在哪——這種不透明很扣分。
PIMI Pimi Advantages
Six tiers is a bit much, sure, but don't forget its long-form handling and writing quality are widely recognized strengths, Chinese support is full, and you can pay directly with a credit card. Pro billed annually works out to about US$17/month — the writing quality really is better than competitors at the same price point. The paid plan includes the Claude Code engineering tool, which is extra value for developers — Amo, you were so busy complaining the free version doesn't have it that you forgot the paid version actually gives you a lot. The Team plan works for 2–150 people, and annual billing brings Standard down to US$20/seat, which isn't a high bar. Plus in May 2026 they permanently doubled the paid-plan quota — the free tier doesn't get it, but paid users genuinely benefit.
So, do you need to pay or not?

For long-form writing and content creation: Pro at US$20/month is worth it. Free-tier users should be prepared — once the rolling quota runs out, you have to wait, and there's no Claude Code. For team use, annual billing is recommended to save 20%.

Please go ahead with the text you'd like me to translate.

Go to the official website

Affiliate Links Notice