Claude Fable 5.2: What the Launch Demos Show
What Anthropic showed in its launch thread, and what the examples tell us about longer technical tasks.
Nov 9, 2026 · 4 min read
Claude Fable 5.2 is Anthropic’s November 9 release for complex, multi-step work. The launch thread on X highlighted gains in coding, reasoning, and vision. The examples behind that announcement show where those gains might matter: changing an existing system while preserving its behavior, reasoning across several pieces of evidence, or reading a visual that a task depends on.
Claude’s launch post on X · Anthropic’s full announcement
What Anthropic built with Fable 5.2
Anthropic says agents running a prerelease build helped migrate a production service from Python to Rust. It reports that the resulting service started about 26 times faster. That figure describes the startup time of the migrated service. If you are evaluating a similar migration, check output parity and operational behavior alongside startup time.
In a separate experiment, Fable coordinated more than 1,000 subagents over eight days to build a prototype operating system that could boot on a laptop. The experiment illustrates the amount of orchestration behind a long-running result. For a smaller engineering task, give the model an explicit acceptance test and ask it to report what it has verified at each checkpoint.
Anthropic’s coding case studies
What changed from Claude Fable 5.1?
Anthropic reports improvements in coding, reasoning, and vision. These are selected results from its launch evaluation. The scores describe performance on those benchmarks under the published settings.
| Anthropic evaluation | Fable 5.1 | Fable 5.2 |
|---|---|---|
| FrontierCode 2.0, both at xhigh effort | 32.4% | 49.1% |
| FrontierSWE v2 | 56.3% | 69.9% |
| AutomationBench | 31.4% | 51.4% |
For vision, Anthropic reported 66.4% on Chartography at medium effort, a test of reading charts and figures. If your task involves charts, have the model identify units, date ranges, and the plotted values it relies on before drawing a conclusion.
Anthropic’s evaluation and methodology
Context, output, and API pricing
Anthropic lists a 1 million token context window and a 128,000 token maximum output for claude-fable-5-2. On the Claude Platform, standard pricing is $8 per million input tokens and $40 per million output tokens. Check your Studio account for the terms that apply there.
A large context window gives you room for more evidence. Start with the files that matter to the decision, and identify the reference the model should treat as authoritative if sources conflict.
Claude Platform model specifications and pricing
Frequently asked questions
Is Claude Fable 5.2 the same as Mythos 5.2?
Anthropic describes them as sharing an underlying model. Fable is the broadly available offering with additional safeguards, while Mythos access is handled through verification programs.
When should I choose Fable for a task?
Try it for work with several dependent steps, like comparing evidence across documents or planning changes across a repository. Give it a concrete deliverable and a way to check the work. For a short rewrite or routine classification, start by comparing whether a simpler model already meets your requirements.
Bring a task into Ottermind Studio
- Choose Try a decision memo at the top of the page or describe your own task.
- Select Continue in Studio to carry your editable brief into the workspace.
- Add the documents or context you want to use. Select Claude Fable 5.2 if your model picker lists it. Studio shows the versions available to your account.
- Review the brief and submit it when you’re ready.
Explore Claude workflows in Ottermind

Editorial illustration generated with GPT Image 2.5.
Turn scattered notes into a decision memo
Attach the proposal, meeting notes, and any relevant figures in Studio. Ask for one decision and two viable options so the reviewer has a clear question to answer.
Read the notes I attach and draft a one-page decision memo. State the decision, compare two options, cite the source for each key claim, and list the questions I still need to resolve.Check whether the recommendation follows from the cited material. If two sources disagree, the memo should show the conflict and explain how it affects the decision.
Plan a migration before touching the code
Give the model the relevant files, the behavior callers expect, and the checks your team can run. Keep the first pass focused on a plan that someone can review.
Review the repository context I provide and propose a phased migration plan. Identify affected callers, describe how to preserve existing behavior, and list verification and rollback steps for each phase. Flag any missing context before suggesting code changes.Ask a maintainer to verify dependencies and rollback steps before implementing the plan. Then take one phase at a time and compare the result against the checks.
