SOOFI Accused of Eval Leakage and Overhype
JJitsev · x · 2026-07-19
The author points out a clear eval leakage in SOOFI's comparisons: its cloned training set has already seen most of the evaluation data, and GPQA even includes the test set. Therefore, claiming it is a "frontier-level champion" is entirely baseless.
He further criticizes the narrative of the SOOFI report: it exaggerates its "transparent" and "sovereign" positioning while failing to clearly state that it was reproduced on Nemotron-3-Nano / the open Nemotron stack. It also uses a self-built "capability index" to support its comparative conclusions, showing obvious packaging issues.
Related event: European Open-Source Model SOOFI Accused of Eval Leakage and Overhype(11 posts)→
More from Models
- Benchmark scores drop from 89% to 19% on new evals — how benchmaxxing breaks leaderboard trust — airesearch12 · 2026-09-11
- ChatGPT tells user their question is too hard and to 'accept dumber answers' — phido3000 · 2026-09-11
- Developer Building a Unified Leaderboard of All Model Benchmark Scores — airesearch12 · 2026-09-11
- Rumor claims Kimi faked performance by serving Claude; DeepSeek new model surprises in evals — realsohamparekh · 2026-09-11
- GPT-5.6 writes well but is instantly forgettable, user complains — BasedRaddka · 2026-09-11
- Opus Refuses Protein Research Codebase Over 'Safety' Concerns, Dev Considers Rolling His Own — josephdviviano · 2026-09-11