On 1 September 2026 Anthropic released two new models: Claude Fable 5.1 and Claude Mythos 5.1. The headlines about it sound like a research lab — benchmarks, science, safety tiers. If you are wondering whether any of this concerns you at all: that is exactly what this article is for. The short answer up front: the real news is not the performance. It is the price.
I will walk through the announcement the way I read it myself: what is new, what it costs, what it means for your plan — and which part you can safely ignore.
What is Claude Fable 5.1?
Claude Fable 5.1 is the successor to Fable 5 — Anthropic's strongest generally available model, built for tasks that run long and independently. The context window stays at one million tokens, still the largest among the frontier models.
And Mythos 5.1? That is the same model — just with different safeguards. Fable 5.1 is the version for everyone, with the usual protections in place. Mythos 5.1 goes exclusively to vetted organizations from cybersecurity and the life sciences through a dedicated access program. That is the part of the announcement making the headlines — “new models aim above all to improve research” — and at the same time the part that has nothing to do with your working day. You can tick it off.
The real news: the same work, roughly a quarter cheaper
The base prices stay unchanged: $10 per million tokens in, $50 out. What Anthropic cut is something else — and drastically: reading from the cache, down from $1 to $0.25 per million tokens. Minus 75 percent.
If “cache” means nothing to you: when Claude works on a task for a long time, it reads the same material many times over — your files, the conversation so far, the instructions. That re-reading runs through a cache, and exactly that is now almost free. The longer and more independently a task runs, the bigger the share of that re-reading on the bill.
That is why Anthropic calculates: for typical usage, total costs drop by around 25 percent — and for heavily agentic work, which is precisely what Fable is built for, by up to 45 percent. A curious aside: reading from the cache now costs half of what it costs on the cheaper Opus 5.
| Price per M tokens (API) | Fable 5 | Fable 5.1 |
|---|---|---|
| Input | $10 | $10 |
| Output | $50 | $50 |
| Cache read | $1.00 | $0.25 |
| Context window | 1M tokens | 1M tokens |
One honest note on this: these prices apply to the developer API — that is, when billing runs by consumption, for instance because an agent works for you or you built something yourself. In the subscription, the monthly price does not change at all. What the plans cost and which one fits which way of working is covered here: What does Claude cost?
Is Claude Fable 5.1 better than Opus 5?
On paper: yes, across the board. Fable 5.1 beats Opus 5 in every benchmark Anthropic has published. Most clearly in scientific work: there the score has more than doubled compared to Fable 5. The coding and tool-use knowledge scores also sit above everything Anthropic has reported before.
And now the sentence you will not find in the headlines: Anthropic itself recommends starting with Opus 5 for most tasks — and only reaching for Fable 5.1 when Opus falls short even at its highest effort level, or when a task really should run for hours without you. Input and output on Fable still cost twice as much as on Opus.
Which means the order from my model comparison stands unchanged: Sonnet for everyday work, Opus for the heavy lifting, Fable only with good reason. The update does not shift the logic — it makes the good reason cheaper.
What does this mean for your plan?
If you use Claude on a subscription, the situation looks like this:
- Max, Team Premium and Enterprise Premium: Fable 5.1 is included. It does count more heavily against your weekly quota than the other models, though — usable up to half of your limits.
- Pro: like Fable 5 before it, Fable 5.1 runs on additional usage credits, not on the regular subscription.
- Free: no access.
There is nothing to reconfigure. Wherever you picked Fable 5 before, you now pick Fable 5.1 — the same model selector in chat, in Claude Code and in Claude Cowork.
One small change you will notice in daily use, however: Anthropic reworked the writing style. Answers come out shorter and less chatty — less preamble, more result.
What you should do now
- Do not rush anything. If you mainly write, plan and analyze with Claude, this update changes nothing for you — your models remain Sonnet and Opus, your subscription price stays the same.
- If you build things or run agents: try Fable 5.1 for your next long task. Same selector, better results on long chains — and if you pay by consumption, the bill shrinks on its own. What such agents can do is in Building an AI agent.
- Remember the pattern. The models get better every few weeks and cheaper to run — without you doing anything. Your job is not to chase every update. Your job is to set up your workflows so that every update reaches you automatically.
The model is the smaller part. The bigger one is what you build with it. That is exactly what we talk about in my free community “KI — aber richtig” — the templates and workflows from my videos live there too. And if you want company instead of doing it alone: the AI Business Community offers monthly live calls, workshops and the MACHZEIT! challenge for exactly that.
Frequently asked questions
What is the difference between Claude Fable 5.1 and Fable 5?
Fable 5.1 delivers better scores in all published benchmarks, writes more concisely and costs less to run: the base prices stay at $10 and $50 per million tokens, but reading from the cache drops from $1 to $0.25 — for typical usage Anthropic expects around 25 percent lower total costs.
What does Claude Fable 5.1 cost?
On the developer API, $10 per million input tokens, $50 for output and $0.25 for reading from the cache. In the subscription, Fable 5.1 is included in Max, Team Premium and Enterprise Premium, but counts more heavily against the weekly limits there; on the Pro plan, access runs on additional usage credits.
What is Claude Mythos 5.1?
Claude Mythos 5.1 is the same model as Fable 5.1, but with different safeguards. It is not freely available — it is granted through an access program exclusively to vetted organizations from cybersecurity and the life sciences. For normal use it plays no role.
Is Claude Fable 5.1 better than Opus 5?
In all benchmarks Anthropic has published, yes. Still, Anthropic itself recommends starting with Opus 5 for most tasks and saving Fable 5.1 for the hardest work — input and output on Fable cost twice as much, and in the subscription it counts more heavily against the usage limits.
Conclusion
Fable 5.1 is a good update with no pressure to act: stronger in the scores, considerably cheaper in continuous operation, same prices, same selector. If you build and run agents, you get more for less. If you write and plan, you do not have to do anything at all.
Which means the answer to the question that comes with every new model stays the same too: it is not the strongest model that wins, but the one whose strength fits your task. Which one that is for you follows from your processes — and that is exactly where my Claude Sprint comes in: five days live, in a small group, and you implement Claude directly in your business.
