Claude Fable 5.2 Review: Faster Tokens, More Thinking

Claude Fable 5.2 is Anthropic's new top model and the successor to Claude Fable 5.1. Most of the launch post covers coding, science, and cybersecurity, but it opens with two changes that apply to every request, faster output and a lower price. Anthropic says the new models "generate outputs up to 30% faster than Claude Fable 5.1." We tested that against Fable 5.1 and Claude Opus 5.5 with the same prompts.

In our tests, Fable 5.2 generated output tokens 35% faster than Fable 5.1, and it was faster on all eight prompts. It also used thinking on every answer, while Fable 5.1 skipped thinking on about half of them. That raised the median wait for the first word from 2.7 to 6.1 seconds and added 22% more output tokens. Across 24 answers, the two models took about the same total time, and Fable 5.2 cost 3% less despite per-token prices that are 20% lower.

What Anthropic Claims

The post doesn't say how speed was measured or at which effort level. Fable 5.2 also has an optional fast mode that Anthropic says "speeds up token generation by up to 2.5 times" at twice the price. Fast mode is a separate setting, so we left it off and tested standard speed only.

Input and output prices drop 20%, from $10 and $50 per million tokens on Fable 5.1 to $8 and $40, according to Anthropic's What's new in Claude Fable 5.2 page. Cache reads drop from $0.25 to $0.20 per million tokens.

The same page lists a behavior change that affects speed. Adaptive thinking is always on, and "at high and above, the model thinks before almost every reply." High is the default effort for both Fable models.

How We Tested

We wrote eight prompts with short, medium, and long expected answers:

• Bug fix. Find the bug in a 12-line binary search.
• SQL query. Return the top 3 customers by 2025 order value in each region, from a two-table schema.
• Debounce function. Write a TypeScript debounce with AbortSignal cancellation and a flush() method.
• Invoice to JSON. Pull the vendor, date, line items, and total from a short invoice.
• Tank word problem. Three pipes fill and drain a tank, one closes partway through, find when the tank is full.
• B-tree explanation. Explain in about 400 words how a B-tree index speeds up range queries in PostgreSQL.
• Callback refactor. Rewrite a callback-based config loader with async/await.
• Technical article. Write a 1,000-word post comparing WebSockets and Server-Sent Events, with code examples.

Each model ran each prompt three times on default settings with no system prompt, for 72 runs in total. Default settings mean high effort for both Fable models and medium effort for Opus 5.5, per Anthropic's effort documentation. For every answer, we recorded when the first part of the response arrived, when the first word of the answer arrived, when the answer finished, and the billed output tokens, which include thinking.

We report output speed as billed output tokens divided by the time from the first part of the response to the last. Thinking time counts toward generation time, and the wait before the response starts doesn't.

Three of the prompts have a single correct answer, and all 27 of those answers were correct. Every model found the hi = arr.length bug, totaled the invoice to $373.00, and found the tank full at 12:12 p.m.

Output Speed

Fable 5.2 was faster than Fable 5.1 on every prompt, by 17% to 48%. The gap was largest on the two long prose answers, the B-tree explanation and the article, where Fable 5.1 slowed to about 70 tokens per second. Across all answers, the gain was 35%, above Anthropic's "up to 30%" figure.

Opus 5.5 generated tokens at about the same speed as Fable 5.2 on most prompts, and faster on the debounce and invoice prompts.

Thinking Before Every Answer

Fable 5.2 used thinking on all 24 of its answers. Fable 5.1 used it on 13. It skipped thinking on the SQL, invoice, and tank prompts in every run, and on the bug fix prompt in two of three. Opus 5.5 used thinking on all 24 at its default medium effort.

The tank problem shows what that costs on a short request:

Pipe A fills a tank in 6 hours and pipe B fills it in 9 hours. Pipe C drains a full tank in 12 hours. All three pipes are opened at 8:00 a.m. with the tank empty, and pipe C is closed at 10:00 a.m. At what time is the tank full?

Both Fable models wrote about 90 words with the same steps and the same result. Fable 5.2 put the answer first in all three runs, while Fable 5.1 put it at the end. Fable 5.1 averaged 250 output tokens and 3.0 seconds per answer. Fable 5.2 averaged 403 output tokens and 6.1 seconds, and 4.1 of those seconds passed before the first word.

The article prompt showed the same pattern on a long answer. Fable 5.2's articles averaged 936 words of prose and 4,191 output tokens. Fable 5.1's averaged 980 words and 2,710 tokens. Fable 5.2 was billed for 55% more tokens for a slightly shorter article, and its first word arrived after about 20 seconds, against 3 seconds for Fable 5.1.

Across all 24 answers, Fable 5.2 wrote 9% less visible text than Fable 5.1 and was billed for 22% more output tokens.

Time and Cost per Answer

The faster tokens and the extra tokens roughly canceled out. The 24 answers took 354 seconds on Fable 5.2 and 360 seconds on Fable 5.1. At Fable 5.1's token counts, Fable 5.2's prices would have brought the total to $1.08. The extra thinking raised it to $1.31, 3% less than Fable 5.1.

Where both models used thinking and wrote a similar number of tokens, Fable 5.2 was faster and cheaper. That covers the debounce, B-tree, and refactor prompts, which finished 15% to 27% sooner and cost 17% to 29% less. On the short prompts where Fable 5.1 skipped thinking, Fable 5.2 took longer, and it cost more on the invoice and tank prompts. On the article, it used 55% more tokens and cost 24% more.

Opus 5.5 cost about half as much as Fable 5.2 across our runs and finished the 24 answers in 14% less time. Anthropic recommends starting with Opus 5.5 for most coding and knowledge work and moving to Fable 5.2 for the hardest problems. On our prompts, all three models were correct wherever we could check.

What to Expect

• Faster token generation. Fable 5.2 generated output 35% faster than Fable 5.1 across our runs, and faster on every prompt.
• Thinking on easy requests. At the default high effort, Fable 5.2 used thinking on every answer, including an invoice extraction and a word problem that Fable 5.1 answered directly. Expect a few more seconds before the first word. Anthropic's docs say a prompt line asking for less thinking won't stop it at medium effort and above, and suggest a lower effort level for easy requests.
• Similar total time. Faster tokens and extra thinking roughly balanced out. Medium-length answers finished sooner, and short answers and the long article took longer.
• Smaller savings than the price cut. Per-token prices are 20% lower, but extra thinking tokens brought the saving on our runs down to 3%. Measure cost per answer on your own prompts instead of comparing rate cards.
• Opus 5.5 for everyday prompts. On these prompts, Opus 5.5 generated tokens at about the same speed (115 against 110 tokens per second), finished sooner, and cost about half as much.

We ran three runs per prompt per model, at default effort, within a few minutes on November 10. Speed varies with load, and our prompts are short, single-turn requests. We didn't test lower effort levels, where Fable 5.2 should use less thinking, or the long agentic sessions Fable 5.2 is built for.

Try It Yourself

Puter.js gives you free access to Claude Fable 5.2 without an API key, straight from the browser. Users cover their own usage costs under the User-Pays Model. This example sends the tank problem, then prints the answer, the wait for the first word, the total time, and the billed output tokens.

Script tag:

<html>
<body>
    <script src="https://js.puter.com/v2/"></script>
    <script>
        const prompt = `Pipe A fills a tank in 6 hours and pipe B fills it in 9 hours. Pipe C drains a full tank in 12 hours. All three pipes are opened at 8:00 a.m. with the tank empty, and pipe C is closed at 10:00 a.m. At what time is the tank full?`;

        (async () => {
            const start = performance.now();
            let firstWord;
            const response = await puter.ai.chat(prompt, { model: "claude-fable-5-2", stream: true });
            for await (const part of response) {
                if (part.type === "text") {
                    firstWord ??= performance.now() - start;
                    puter.print(part.text);
                }
                if (part.type === "usage") {
                    const total = performance.now() - start;
                    puter.print(`<br><br>First word: ${(firstWord / 1000).toFixed(1)} s, total: ${(total / 1000).toFixed(1)} s, output tokens: ${part.usageDetails.outputTokens}`);
                }
            }
        })();
    </script>
</body>
</html>

npm:

// npm install @heyputer/puter.js
import { puter } from "@heyputer/puter.js";

const prompt = `Pipe A fills a tank in 6 hours and pipe B fills it in 9 hours. Pipe C drains a full tank in 12 hours. All three pipes are opened at 8:00 a.m. with the tank empty, and pipe C is closed at 10:00 a.m. At what time is the tank full?`;

const start = performance.now();
let firstWord;
const response = await puter.ai.chat(prompt, { model: "claude-fable-5-2", stream: true });
for await (const part of response) {
    if (part.type === "text") {
        firstWord ??= performance.now() - start;
        process.stdout.write(part.text);
    }
    if (part.type === "usage") {
        const total = performance.now() - start;
        console.log(`\n\nFirst word: ${(firstWord / 1000).toFixed(1)} s, total: ${(total / 1000).toFixed(1)} s, output tokens: ${part.usageDetails.outputTokens}`);
    }
}

To compare, run the same prompt with claude-fable-5-1 or claude-opus-5-5. Run it a few times, since timing and token counts vary between runs.

FAQ

What is Claude Fable 5.2?

Claude Fable 5.2 is Anthropic's top generally available model, built for demanding reasoning and long-running agent work, according to Anthropic's model overview. It replaces Claude Fable 5.1, and adaptive thinking is always on. Anthropic recommends starting with Claude Opus 5.5 for most work and moving to Fable 5.2 when Opus 5.5 doesn't meet your requirements.

How much does Claude Fable 5.2 cost?

Claude Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, 20% less than Fable 5.1's $10 and $50, per Anthropic's What's new page. Cache reads cost $0.20 per million tokens, batch requests cost half, and fast mode costs $16 and $80 per million tokens. In our tests, Fable 5.2 used 22% more output tokens than Fable 5.1, so it cost only 3% less across the same 24 requests.

How fast is Claude Fable 5.2?

Claude Fable 5.2 generated about 110 output tokens per second in our tests, 35% faster than Fable 5.1's 82. Anthropic says it's up to 30% faster than Fable 5.1, and its optional fast mode speeds up generation by up to 2.5 times. At the default high effort, Fable 5.2 used thinking on every answer, so its median wait for the first word was 6.1 seconds, against 2.7 seconds for Fable 5.1.

Is Claude Fable 5.2 better than Claude Fable 5.1?

Claude Fable 5.2 scores higher than Fable 5.1 on Anthropic's benchmarks, for example 49.1% against 32.4% on FrontierCode 2.0 at xhigh effort, according to the launch post. In our tests, it generated tokens faster but used thinking on more requests, so total time across 24 answers was about the same. Anthropic's What's new page lists four changes from Fable 5.1 that can break existing code, covering thinking blocks, the computer use tool, and the advisor tool.

What is the difference between Claude Fable 5.2 and Claude Mythos 5.2?

Claude Fable 5.2 and Claude Mythos 5.2 are the same underlying model, with the same specs and pricing. Fable 5.2 has cybersecurity and biology safeguards and is available to all customers. Mythos 5.2 is available only to organizations verified through Anthropic's verification programs, such as the Cyber Verification Program.

What is the Claude Fable 5.2 context window?

Claude Fable 5.2 has a 1-million-token context window and 128,000 max output tokens. It accepts text and image input, returns text, and has a June 2026 knowledge cutoff, according to Anthropic's model overview.

Where can I use Claude Fable 5.2?

Claude Fable 5.2 is available on the Claude API as claude-fable-5-2, on Amazon Bedrock, Google Cloud, and Microsoft Foundry, and in the Claude apps on paid plans. Developers can also call it through Puter.js with the model ID claude-fable-5-2.

Related

• Claude Fable 5.1 Review: Better Results with Less Effort
• Claude Opus 5.5 Review: Fable-Level Results for Less
• Kimi K3.1 Pro Review: Careful Answers, Longer Reasoning
• Free, Unlimited Claude API
• Anthropic Claude API pricing
• Anthropic's official Claude Fable 5.2 launch post
• What's new in Claude Fable 5.2
• Puter.js AI chat() reference