Running ops with Fable 5 + Adapt has been insane

As engineers in SF, we've all been anticipating the release of Mythos models, the model apparently so powerful that the Anthropic Red Team wouldn't allow for its release.

That's why when Anthropic released Claude Fable 5 (a Mythos-class model) yesterday, we had the instant urge to switch it on within Adapt and see for ourselves. We spent yesterday afternoon switching over to it and testing the model out internally. The results are quite remarkable.

We've seen a jump in the quality of Adapt's responses, generation and interpretation of images, and writing code for Adapt apps. It's tremendously helped our users who connect across all their apps and run company operations through Adapt's agent. Here are a couple of pretty impressive tasks we've been able to run with Fable 5.

What is Adapt

For context, we built Adapt to be an agent specialized in unifying your business knowledge and running operations for you: your company's agent. It works by connecting across all of your apps and integrations. It can help you answer questions and do actual work for you.

For example, you can ask Adapt “What is my ad spend for the month?” and Adapt will look across all of your ads platforms like Meta, LinkedIn, Google Ads to compile a blended spend. The best part is that people, who are not in the marketing department, can ask these questions. Everyone has cross-department knowledge.

Fable 5 can build insane Adapt apps in one prompt

The first thing we wanted to do was test Fable 5's ability to create Adapt apps. Adapt apps are internal apps you can build for your companies, such as custom dashboards that visualize company data. We host Adapt apps for you, and they are accessible by anybody in the company.

One of the coolest apps we built on Fable 5 in Adapt is a real-time earthquake tracker that uses the real-time USGS earthquake API. Fable 5 was able to cleverly search for 3D rendering JavaScript libraries out there, and a globe visual with real time earthquake analytics. This was all done in a single Adapt prompt that took less than 5 minutes.

Anthropic's announcement post did say agentic coding improved the most, jumping from 69.2% in the agentic coding benchmark with Opus 4.8 to 80.3% with Fable 5. Typically, benchmarks don't mean much, but the improvement gap this time is certainly felt in Adapt.

Fable 5 supercharges knowledge work

We've also noticed a huge jump in Adapt's ability to do knowledge work when running Fable 5. Compared to when we were using Opus 4.8, Adapt's accuracy in reasoning over documentation and deep research has significantly improved.

This morning we gave Fable 5 running on Adapt an excel spreadsheet for our expenses and financials, and access to the Ramp integration, to forecast our expenses and books for the month of June. Adapt created a forecast report with precise and accurate numbers.

For research on this article, we asked Fable 5 to do deep web research about itself and comparisons to previous Claude models. Adapt did the web research, provided the sources, and generated a PDF report breaking down the full analysis. This was all done within Slack.

Our favorite part of doing knowledge work on Adapt is that everything lives in Slack, and the knowledge is shared with everyone. The whole team has read the report about Fable 5, and we're now all aligned on its capabilities.

See Fable 5 in Adapt

Fable 5 is expensive to run directly through other platforms and APIs. Qualified businesses can evaluate it through a guided Adapt pilot built around their own tools and workflows.

If this sounds interesting, get a demo with our team.