Claude Fable 5.1 Review: Better Results with Less Effort

Claude Fable 5.1 is out. It's Anthropic's top-tier model, above Opus, Sonnet, and Haiku, and the update is built around holding the same quality at lower reasoning effort than Fable 5 needed. Since effort maps directly to cost, we put together an estimate of what that means for coding work specifically.

What changed

Anthropic's launch post describes Fable 5.1 as reaching similar or better results than Fable 5 at low or medium effort, with clearly higher results at higher effort tiers. We can't grade output quality ourselves for this post, so instead we worked out the pricing behind that claim and estimated what it's worth in dollars for a coding workload.

Pricing

Fable 5.1 keeps the same input and output pricing as Fable 5, $10 per million input tokens and $50 per million output tokens. The change is in cache reads, which drop from $1.00 to $0.25 per million tokens, a 75% cut. Everything else stays the same: 1,000,000 tokens of context and a 128,000-token max output on both models.

What "same result, less effort" adds up to

Two different things get cheaper here, and they stack.

First, cache reads cost 75% less on Fable 5.1. Agentic coding sends a lot of repeated context (files, tool definitions, conversation history) as cache reads every turn, so this alone moves the total bill. Anthropic's own estimate is that this brings typical workloads down about 25%.

Second, and this is the part that's easy to miss if you only look at the pricing table. If Fable 5.1 genuinely gets the same or better result at low or medium effort instead of Fable 5's default high effort, it's spending fewer reasoning tokens to get there. Less effort means less thinking, which means fewer tokens billed, on top of the cache-read savings. Anthropic hasn't published a number for how much smaller that reasoning-token spend is per effort tier, so we can't quantify this part from their materials. It's a separate lever from caching though, and for a reasoning-heavy coding task it could matter more than the cache-read discount does.

To see roughly how the two effects stack, we ran the numbers on an illustrative session, adding our own assumption for the effort-tier token reduction on top of the confirmed cache pricing.

Assumptions:

• 100 turns in a session
• 50,000 cache-read tokens per turn (same context-reuse pattern on both models)
• 500 fresh input tokens per turn (effort level doesn't change how much you send in)
• Fable 5 at its default high effort: 800 output/reasoning tokens per turn
• Fable 5.1 at medium effort: roughly a third fewer output/reasoning tokens per turn to reach the same result, about 530 tokens per turn. This is our own assumption for illustration, not a figure Anthropic has published.

Totals across the session: 5,000,000 cache-read tokens and 50,000 fresh input tokens on both, 80,000 output tokens on Fable 5, and 53,000 output tokens on Fable 5.1.

Cost on Fable 5 (high effort):

• Cache reads: 5M × $1.00/M = $5.00
• Input: 0.05M × $10.00/M = $0.50
• Output: 0.08M × $50.00/M = $4.00
• Total: $9.50

Cost on Fable 5.1 (medium effort):

• Cache reads: 5M × $0.25/M = $1.25
• Input: 0.05M × $10.00/M = $0.50
• Output: 0.053M × $50.00/M = $2.65
• Total: $4.40

By this estimate, the combined effect of cheaper cache reads and lower reasoning spend brings the session cost down by about 54%, versus roughly 39% from the cache-read discount alone. That's above Anthropic's own 25% figure for typical workloads, which only accounts for the caching side since that's the part that follows directly from the price sheet. The reasoning-token side depends on how well Fable 5.1 actually holds up at lower effort, which is a separate question from anything in the pricing table.

The obvious caveat: both numbers move a lot depending on the actual session. A short chat-style exchange with little repeated context and little reasoning won't see anywhere close to this. A long-running coding session with a large repo loaded into context and genuinely hard reasoning steps will see more of both effects, since more of the total spend sits in the columns that got cheaper.

Anthropic describes Fable 5.1's coding behavior as carrying more of a project on its own across multi-hour sessions, being less likely to disable a failing test to get to green, and being more willing to say when it's stuck rather than guessing. Combined with the pricing above, that points at long agentic coding sessions as the case where both the cost curve and the described behavior changes would show up the most.

Try it yourself

Puter.js gives free access to Fable 5.1 without an API key, so you can try the effort settings yourself.

Script tag:

<html>
<body>
  <script src="https://js.puter.com/v2/"></script>
  <script>
    puter.ai.chat("Write a function that debounces API calls with cancellation support", {
      model: "claude-fable-5-1",
      reasoning_effort: "medium"
    }).then(response => {
      puter.print(response.message.content[0].text);
    });
  </script>
</body>
</html>

npm (@heyputer/puter.js):

import { puter } from '@heyputer/puter.js';

const response = await puter.ai.chat(
  "Write a function that debounces API calls with cancellation support",
  { model: 'claude-fable-5-1', reasoning_effort: 'medium' }
);

console.log(response.message.content[0].text);

Puter.js also exposes reasoning_effort values of none, minimal, low, high, and xhigh if you want to compare effort levels yourself. Response usage data is available on the returned object if you want to track token spend per call.

FAQ

When was Claude Fable 5.1 released? September 1, 2026, alongside Claude Mythos 5.1.

What's the pricing? $10 per million input tokens and $50 per million output tokens, unchanged from Fable 5. Cache-read pricing drops from $1.00 to $0.25 per million tokens, a 75% reduction.

What's the context window? 1,000,000 tokens of context with a 128,000-token maximum output, the same as Fable 5.

What actually changed compared to Fable 5? Three things, according to Anthropic: pricing (cheaper cache reads), safeguards (60% fewer cybersecurity false positives, and Fable 5.1 can now be used to discover software vulnerabilities though not develop exploits for them), and capability (higher performance on hard reasoning benchmarks, plus improvements to long-running agentic coding and autonomous operation across multiple applications).

Is Fable 5.1 the same model as Mythos 5.1? Yes. Anthropic describes them as the same underlying model with different levels of safeguards. Fable 5.1 is generally available. Mythos 5.1 is limited to Anthropic's trusted access programs, with safeguards aimed at cybersecurity and life sciences use.

Related

• Anthropic: Introducing Claude Fable 5.1 and Claude Mythos 5.1
• Claude Platform docs: Effort levels
• Puter.js AI documentation