Claude Fable 5.2 vs 5.1: 20% cheaper per token, not always per task

Claude Fable 5.2 does the same work as Fable 5.1 for less money per token and, according to Anthropic, writes its answers faster. What it does not promise is a smaller bill. The cost of a task depends on how many tokens the model spends thinking, and an early independent test found the new version spending far more of them.

That gap matters to anyone choosing between the two versions in Claude Code, in the Claude apps or through the API. Anthropic has cut the list price by a fifth and points to large jumps on coding and agent benchmarks. Developers who simply swap the model name should look at their effort setting before they look at the price sheet.

Price: a fifth off every token

Fable 5.2 costs $8 per million input tokens and $40 per million output tokens. Fable 5.1 cost $10 and $50, according to Anthropic’s own pricing page, so both rates fall by 20%. Reading cached context drops from $0.25 to $0.20 per million tokens, and batch jobs, billed at half price, now cost $4 and $20. Claude Mythos 5.2, the same underlying model without some of Fable’s cybersecurity and biology safeguards, carries the identical price.

A new optional fast mode generates tokens up to 2.5 times faster, Anthropic says, for double the rate: $16 per million input tokens and $80 per million output tokens. It works in Claude Code, on the Claude Platform, on Google Cloud and in Microsoft Foundry. Fast mode for Mythos is limited to organizations inside Anthropic’s verification programs.

Speed and scores: where 5.2 pulls ahead

Anthropic says Fable 5.2 produces output up to 30% faster than Fable 5.1. Computing for Geeks, which ran both models on the same three DevOps tasks, measured 17% to 25% faster streaming at medium effort. That is short of the headline figure, but you notice it. The context window stays at one million tokens, with up to 128,000 tokens of output, the site reports.

The benchmark gains are bigger. On FrontierCode 2.0, which checks whether open-source maintainers would merge a model’s code changes, Fable 5.2 scores 49.1% at its Xhigh effort setting against 32.4% for Fable 5.1. On the Remote Labor Index, which measures how much real freelance work a model can complete, it reaches 41.3% against 17.9%. Humanity’s Last Exam with tools rises from 65.6% to 74.2%, and the RoboDojo robot-control simulation jumps from 27.6% to 64.9%. These are Anthropic’s runs, at each model’s highest available effort unless noted.

An outside ranking points the same way. Artificial Analysis puts Fable 5.2 at maximum effort first on its Intelligence Index, with a score of 65, ahead of Gemini 4 Carbon at 60 and Claude Opus 5.5 at 59. The same listing ranks it last of 73 models on cost per index task, at $26.54, and calls it very verbose: it generated 275 million tokens to finish the test suite, against a median of 72.8 million.

Fewer refusals are the quieter upgrade. Anthropic reports about 55% fewer safety blocks than with Fable 5.1, roughly 60% fewer in cybersecurity and 40% fewer in biology. In the Computing for Geeks rerun, Fable 5.1 blocked two of eight routine sysadmin requests, including a port scan of the tester’s own network, while Fable 5.2 answered all eight in full.

The catch: effort decides the bill

Per-task cost is where the upgrade gets complicated. At medium effort, the default in Claude Code and the Claude apps, Fable 5.2 wrote 10,938 tokens across the three Computing for Geeks tasks, where Fable 5.1 wrote 5,510. Most of the difference was thinking, which bills at the output rate. So a token 20% cheaper produced a bill 56% bigger, $0.446 against $0.285. The code was no more correct, although Fable 5.2’s files were more thorough, with extra validation checks and resource limits.

The test has clear limits. It used short, one-shot prompts, mostly single runs through OpenRouter, and only three tasks, while Anthropic built Fable for long agentic sessions and says that at medium effort Fable 5.2 completes challenging tasks with fewer tokens than Fable 5.1. Anthropic’s effort sweep on FrontierCode, as reproduced from its system card by Computing for Geeks, backs the wider picture: at high effort Fable 5.2 scores 45.3% for about $6.72 per task, while Fable 5.1’s best result of 32.4% cost $8.68. Every benchmark figure in that comparison still comes from the company selling the model.

For teams moving from 5.1, the sensible approach is to watch spending in the first week and set effort per job rather than once. According to Anthropic, Fable 5.2 at high effort roughly matches the best FrontierCode score of Opus 5.5, 45.3% against 45.5%, for about half the cost per task, and the company still recommends starting most coding and knowledge work on Opus 5.5. Developers whose code edits conversation history have one more thing to check: Anthropic’s preserved thinking protection now applies to every account, and Computing for Geeks reports that replayed reasoning which no longer matches its original context now returns an error.

Claude Fable 5.2 launched on November 9, 2026, replacing Fable 5.1, which arrived in September. It is available in the Claude apps on eligible paid plans and to developers as claude-fable-5-2 on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure, in the countries where Anthropic sells Claude, a list that leaves out mainland China, Hong Kong and Russia. During launch week, eligible paid plans get one to three Fable usage resets depending on how long they have subscribed, and Fable can take up to half of the weekly limit on Pro plans and standard Team seats; Computing for Geeks reports that the Pro and Team offer runs until November 16. Mythos 5.2 is open only to organizations accepted into Anthropic’s Cyber Verification and Life Sciences Verification programs, and to Claude Enterprise customers through Claude Security.

For a sense of what the extra headroom can do, one Anthropic engineer had Fable 5.2 coordinate more than 1,000 subagents for eight days to build a prototype operating system from an empty folder. It boots on a laptop. Anthropic has not said what that run cost.

Tags: artificial intelligence, Anthropic, Claude, AI models, Claude Fable 5.1, Claude Fable 5.2, Claude Mythos 5.2