Claude Fable 5 vs. GPT-5.5 Pro: 12 Tasks Compared Through the BotHub Aggregator

The authors of the BotHub blog on Habr compared two top-tier language models, Anthropic's Claude Fable 5 and OpenAI's GPT-5.5 Pro, on 12 practical tasks. They tested both through their own neural network API aggregator.

Why this particular pair

In an earlier article, the authors compared Claude Opus 4.8, GPT-5.5 and Gemini 3.1 Pro, but noted up front that GPT-5.5 Pro was left out: it is a fairer match for Claude Fable 5, which was unavailable at the time of publication. That changed when Anthropic unexpectedly returned Claude Fable 5 to general availability.

• Claude Fable 5 is the first model in Anthropic's new Mythos class, which sits above Opus in the company's hierarchy
• Anthropic unexpectedly made Fable 5 publicly available again
• The comparison is built on 12 practical, everyday tasks rather than formal benchmarks
• Testing was done through the BotHub API aggregator, not through the models' web interfaces

How the testing method works

The authors deliberately passed on ready-made benchmarks and vendors' marketing claims, betting instead on real everyday tasks. Testing ran through the BotHub neural network aggregator, which works via API. According to the authors, this strips out the crutches and workarounds that implicitly help models in their branded web interfaces. The method also shows right away how much it costs to solve each task.

Cost is measured in BotHub's internal currency, CAPS, tied to the number of tokens spent. Normally one ruble buys between 4000 and 6500 CAPS depending on purchase volume; in this comparison the authors used a rate of about 1 ruble = 6370 CAPS, because their plan changed. Both models belong to the premium segment, so the final per-task prices came out noticeably higher than in the earlier comparison of Opus 4.8, GPT-5.5 and Gemini 3.1 Pro.

As before, the authors pick the winner of each task mostly subjectively, and they say so openly. All testing materials are publicly available, so readers can form their own opinion and choose their favorite among the models.

What makes this format interesting

Publishing all the testing materials is part of the authors' method: rather than simply declaring a winner, they let readers judge both models' answers to each of the 12 tasks and reach their own conclusions, even if those differ from the authors'. This approach is especially valuable for premium models such as Claude Fable 5 and GPT-5.5 Pro, where the cost of a request becomes part of the decision: not only the quality of the answer, but its price in CAPS.

What this means

With Claude Fable 5 available again, it is now possible for the first time to compare Anthropic's top Mythos-class model directly with the top Pro version of OpenAI's GPT-5.5 on equal API terms, rather than relying on developers' marketing promises. For teams choosing a model for specific tasks, practical comparisons with transparent token-based cost accounting are worth more than formal benchmarks, which premium models usually pass almost equally well.

Frequently asked questions

What is Anthropic's Mythos model class?

Mythos is a new class of Anthropic models that sits above Opus in the company's hierarchy. Claude Fable 5 is the first model released in this class.

How was the cost of model requests calculated in the comparison?

Cost was measured in the BotHub aggregator's internal currency, CAPS, tied to the number of tokens spent. Normally 1 ruble equals 4000-6500 CAPS depending on purchase volume; this test used a rate of about 1 ruble = 6370 CAPS.