Strawberry Tabby Leak Evidence: Analyzing the Claims, Documents, and Data
The controversy began when an anonymous repository appeared on an encrypted file-sharing server, containing 4.2 gigabytes of structured JSON data, internal benchmark sheets, and slide decks detailing internal testing runs. The title of the bundle mashed together "Strawberry", the long-running project code name for advanced mathematical reasoning models, with references to financial reporting spearheaded by Tabby Kinder, whose reporting on tech market valuations and massive capital rounds had circulated among venture circles.
The documents quickly moved from private Discord servers to public X feeds and Reddit discussion boards. Within forty-eight hours, specialized machine learning communities began running replication scripts against the published outputs. Unlike generic hype dumps that frequently plague developer forums, this release carried specific test-run identifiers, API endpoint paths, and model parameter notes that matched proprietary architectures deployed throughout late 2024 and 2025.
The initial public reaction split between panic and skepticism. Enterprise customers worried that custom prompt data or fine-tuning datasets had been exposed to the public. Meanwhile, competing AI labs raced to dissect the performance figures to determine whether the documented leap in complex logic tasks was reproducible or merely an artifact of cherry-picked test prompts.
