Testing Claude Fable 5: A Financial Analysis Review

On June 9, 2026, Anthropic released Claude Fable 5, a Mythos-Class model with advanced safeguards in some areas such as cybersecurity, distillation, chemistry and biology.

Anthropic has published numbers showing Fable 5 is ahead of Opus 4.8 and GPT-5.5 on several key benchmarks. In GDPval-AA (knowledge work), it is significantly ahead of Opus 4.8, GPT 5.5 and Gemini 3.1 Pro.

According to Claude, "Fable 5 shows strong performance on complex analytical tasks.".

We decided to test Fable 5 by running an analysis of 'Space Exploration Technologies Corporation(SpaceX)' for investment research purposes, using the 'comps-analysis' skill. The 'skill.md' file was from Anthropic's financial-services GitHub repository.

The task took around 25 mins. Claude did not ask any clarifying questions except to request permission to access external websites. Impressively, Fable 5 was capable of finding the peer companies for comparison.

*Please note: This is for research and educational purpose and does not constitute investment advice. Original generated files can be downloaded here, SpaceX_Comps_Analysis_2026-06-10.xlsx, and, SpaceX_Comps_Memo_2026-06-10.pdf.

When we tested the same skill using Qwen deep research, we had to manually enter the peer companies. The numerical figures are differed slightly compared to Fable 5.

*The Qwen-generated report can be downloaded here, The_SpaceX_Comp_Analysis_A_Statistical_Benchmark_of_Growth_Efficiency_and_Valuation_Against_Public_Peers.pdf.

Although both reports from Claude and Qwen were produced based on the same methodology specified in the skill, the numbers varied because Claude and Qwen pulled from different sources. For example, Claude Fable 5 used Stock Analysis while Qwen used SEC.gov.

The report from Qwen also contained much more information regarding the business model, capital structure and operation.

After comparing reports, we have decided to try again with /research command in Claude desktop app.

The newly generated analysis report, SpaceX_Analysis.pdf, included extra recommendations and caveats sections. Compared to Qwen deep search, it still lacked detailed company information such as operation and capital structure.

Is Fable 5 worth the premium price tag? It might be for hard, long tasks and autonomous coding. But for certain knowledge work, like company analysis, it may not be the best value yet.

At then end, we tested how well Fable 5 performs in translation. Strangely, the model auto switched back to Opus 4.8.

Does this mean any long-context conversion will be reverted back to Opus 4.8 ? Is this just a temporary quirk during the usage before 22nd June? Or is the system simply smart enough to tell the translation doesn't require such a powerful model.