# Claude Fable 5.2 launches on Devin with higher benchmark scores

> The new model tops FrontierCode 2.0 at 49.1% and cuts cost per task by more than two‑thirds.

Oossa · 2026-11-09 · https://oossa.com/en/claude-fable-5-2-launches-on-devin-with-higher-benchmark-scores

Anthropic released Claude Fable 5.2 on November 9, 2026, and made it available through the Devin platform. Devin offers a desktop app and a command‑line interface, so developers can start using the model right away. The upgrade replaces the earlier 5.1 version that many teams were already testing on real‑world engineering problems.

The company says the model was evaluated on FrontierCode 2.0, a benchmark that simulates typical software‑engineering tasks. The benchmark scores each solution on quality, how easy it is to merge, and whether it passes basic safety checks. Fable 5.2 earned a weighted score of 49.1%, beating its predecessor’s 32.4% and also outpacing the competing Opus 5.5 model, which scored 45.5%.

Beyond the headline number, Anthropic notes that Fable 5.2 reaches the best result of the older model while using less than a third of the average dollar cost per task. Cost is measured at the list API price, after any public discounts. This means teams can get higher‑quality code suggestions without spending as much on API calls.

The rollout is immediate for anyone who downloads the Devin Desktop client or installs the Devin CLI. No special licensing is mentioned, and the model is accessed through the standard Devin pricing plan.

## How the benchmark works

FrontierCode 2.0 asks a model to write or modify code, then checks the output against a set of rules. If the answer fails a blocking criterion—like producing unsafe code—it gets a zero for that run. For runs that pass, the rubric grades the solution on readability, correctness, and how easily it could be merged into an existing codebase. The final score is a weighted average of those items.

Anthropic reports that the cost figure includes the mean USD spend per rollout at the public list price, after discounts. Because Fable 5.2 achieves similar or better scores with lower spend, developers can run more iterations within the same budget.

## The facts

- Claude Fable 5.2 became publicly available on Devin on 2026-11-09.
- On the FrontierCode 2.0 benchmark, Fable 5.2 scored 49.1%, up from 32.4% for version 5.1.
- Fable 5.2 outperformed Opus 5.5, which scored 45.5% on the same benchmark.
- The new model reaches the best result of Fable 5.1 while using less than one‑third of the average cost per task.
- Developers can download the model via Devin Desktop or install it with the Devin CLI.

## Why it matters

For software teams, the upgrade means faster, higher‑quality code suggestions without a proportional rise in API costs. Smaller firms can experiment more with AI‑assisted coding, while larger groups may reduce their cloud spend. Independent validation of the benchmark results has not been published yet.

## Sources & references

1. [Claude Fable 5.2 is now available in Devin](https://windsurf.com/blog/claude-fable-5-2) – Windsurf, 2026-11-09

Last updated: 2026-11-09