TL;DR: On July 18, 2026, Anthropic reversed its month-long plan to move Claude Fable 5 onto metered pricing and is keeping it in subscriptions permanently. From July 20, Max and Team Premium plans include it at 50% of their limits; Pro and Team Standard reach it through usage credits plus a one-time $100 credit. The reversal follows GPT-5.6 Sol on July 9 and Moonshot’s cheaper Kimi K3 in mid-July — competition that made a subscription without the vendor’s best model hard to sell.

For a month, the plan was to take Claude Fable 5 out of subscriptions. Anthropic released it on June 9, pulled access on June 12, restored it on July 1, and kept telling subscribers the same thing: use it while it is bundled, because when the promotional window closes it moves to metered, usage-credit pricing. That window kept sliding — from July 7 to July 12 to July 19, each extension announced close to the deadline it replaced. On July 18, a day before the latest cutoff, Anthropic announced it was keeping Fable 5 in the subscription plans permanently.

What the competition did to the plan

The reversal reads more clearly against what shipped in the ten days before it. OpenAI put out GPT-5.6 on July 9, with Sol as its flagship. Moonshot AI followed in the middle of the month with Kimi K3, a roughly 2.8-trillion-parameter model. By the company’s own account and early independent testing, K3 still trails Fable 5 and Sol on overall performance, but it beats the tier just below them — Claude Opus 4.8 and GPT-5.5. What stands out is the price: Kimi K3 costs a fraction of what the American frontier models charge, and CNBC’s reporting framed exactly that cost gap as pressure labs like Anthropic and OpenAI now feel from cheaper models on the market. The slow arrival of models that do most of the same work for less is a pattern I have looked at before on the open-weight side.

Simon Willison read the reversal as plain competitive pressure, and put the subscriber’s question bluntly: why pay $100 or $200 a month for a plan that does not include the vendor’s best model?

What changes on July 20

For subscribers the practical picture is short. Starting July 20, Max and Team Premium plans include Fable 5 at 50% of their plan limits. Pro and Team Standard plans keep access through usage credits — the metered route the whole subscriber base had been headed toward — and get a one-time $100 credit to start. Metered here means the API rate the model has carried since launch: $10 per million input tokens and $50 per million output — the most expensive tier Anthropic sells, and its most capable. The top plans fold Fable 5 into the flat fee at reduced limits, while Pro and Team Standard meter it with the credit softening the first stretch.

This kind of back-and-forth, and the competition driving it, tends to land in the customer’s favour. Rival labs pushing each other, and a vendor willing to keep revising its own terms in public, is uncomfortable to sit through.

I have been living the same month as anyone on these plans — I wrote about day one back in June, and the weeks since have been one reversal after another: the model right there to use, and never quite safe to count on. If you spent that month with Fable 5 too, I would like to hear how it went — how much more did you actually get done with it?

TL;DR：Anthropic 原本打算在免費期結束後把 Claude Fable 5 從訂閱方案移除、改走計量付費，這一個多月截止日一延再延。7/18 方向翻轉：7/20 起 Fable 5 併回訂閱，所有 Max 跟 Team Premium 方案直接含（用到方案額度的一半），Pro 跟 Team Standard 走 usage credits 外加一次性 $100。推這一把的是競爭，GPT-5.6 Sol 跟 Kimi K3 這半個月接連落地。這種你來我往加上廠商反覆調整，最後受益的是用的人。也想聽你這一個多月用 Fable 5 的心得。

六月我寫第一天用 Fable 5 的心得時，結尾是這樣交代訂閱的：6/22 之前 Pro、Max、Team 方案含著 Fable 5，6/23 起移除、改走 usage credits，token 帳等免費期結束再重算。

結果短短時間內就有很大的變化。

一個多月，截止日一延再延

先是模型本身。Fable 5 中間一度整個下線，隔了一陣子才恢復。回來之後免費期跟著重開，官方給了個 7/7 的截止日：時間到那天當天，宣布延到 7/12；7/12 過了，再延到 7/19。每次都是快到期才宣布再給幾天。對真的在排預算的訂戶來說，貴不貴還好說，真正難的是不知道哪天要開始另外算錢，一直沒個定數，但是又有額度可用，快樂痛苦並存著的感覺。

如果往後用 Fable 5 是照 API 牌價計量：每百萬 input token $10、output $50，在 Claude 家族中是最貴也最好的等級。免費期內錢包沒感覺，一旦改計量，長任務跟掛機那種用法會讓 token 瞬間蒸發，這也是六月那篇說要重算的意思。

7/20 起，Fable 5 留了下來

7/18 這天，方向反了過來。Anthropic 沒有按原訂計畫把 Fable 5 推去計量，而是宣布 7/20 起把它併回訂閱：所有 Max 跟 Team Premium 方案直接含 Fable 5，可以用到方案額度的一半；Pro 跟 Team Standard 仍走 usage credits，但額外補一筆一次性的 $100 credit。

對照六月那篇的擔心，這是最好的一種轉彎：原本大家要擔心 Fable 5 的額外消耗，有新的解法了。Max 訂戶還是原價，多了一個 Mythos 級模型可以用到額度的一半。

為什麼會這樣轉

推這一把的不只 Anthropic 自己。OpenAI 7/9 放出 GPT-5.6，旗艦叫 Sol；一週後，Moonshot 推出 Kimi K3，2.8 兆參數。目前的評測裡它整體還排在 Fable 5 跟 Sol 之後，價錢卻壓低一截。

Simon Willison 把這次轉向讀成競爭逼的：當對手都把最強的模型直接放進訂閱，你的訂閱方案裡卻要另外收費才能用最強的，商業上撐不久。他還多猜了一句，Anthropic 可能得挪一些原本訓練用的 GPU 去撐服務量，才吃得下把 Fable 5 開給所有 Max 訂戶的用量（這句是他的推測）。

Kimi K3 那邊的效應更直接。它用明顯低的價錢逼到最前緣邊上，把一個問題推到檯面上：最前緣的模型還能賣這個價錢多久？Anthropic 選擇不收回、還加碼 $100，某種程度就是在回答這個問題。

這種來回，對使用者是好事

把這一個多月看下來，這裡面沒有誰特別佛心。要不是 Sol 跟 K3 在同一個月擠進來，7/20 這個決定不見得會長這樣。價格跟供給被同業逼著往下走，對用的人來說就是最實在的那種好處，這回剛好看得很清楚。

想聽你的心得

免費期一路延到現在，應該不少人這一個多月都在拿 Fable 5 跑東西。你的體感如何：值不值它的份量、哪類任務丟給它最有感、跟 GPT-5.6 Sol 或 Kimi K3 比起來又怎樣？有多完成多少任務呢？

延伸閱讀

• Claude Fable 5 是什麼？第一個公開的 Mythos 級模型，加上我第一天的使用心得：這篇的六月版，Fable 5 是什麼、定價、跟 Opus 的關係都在那
• GPT-5.6 的 Sol、Terra、Luna 是什麼：這次逼 Anthropic 轉向的對手之一，三個變體差在哪
• OpenRouter 上，美國公司用中國模型的 token 佔比升到每週三成以上：Kimi 這類模型用低價往上打的大背景
• Anthropic Reverses Course: Claude Fable 5 Stays in Subscription Plans：同主題英文版