Model Accused of Architecture Copying and Benchmark Contamination
eliebakouch · x · 2026-07-16
This post criticizes a "sovereign" foundation model:
- The model is accused of having nearly the same architecture as Nemotron 3 Nano and 80% identical data mix; "sovereign" is seen as overhyped.
- More seriously, the author suggests they seem to have included GPQA Diamond benchmark content in training data.
- A specific example: their dataset AIML-TUDA/QA-base is only lightly paraphrased from GPQA Diamond questions, like "training on eval set."
- The author questions whether a significant part of the model's capability gap comes from contamination/leak training rather than better training strategies.
Related event: SOOFI benchmark claims challenged over leakage and baseline reporting(13 posts)→
More from Models
- Kimi K3.1 Pro regresses on Debate Benchmark while DeepSeek 1017 gains at lower cost — teortaxesTex · 2026-10-27
- 4 Hours of LLM Startup Research Produced a Convincing but Nonsense Product Thesis — annetgriffin · 2026-10-27
- Dev argues Claude burns far more tokens than GPT models for the same coding work — haider1 · 2026-10-27
- MiniMax matches Claude Opus 4.5 in non-benchmaxxed evals, users report — teortaxesTex · 2026-10-27
- Real-world math: GPT-6.1 Terra's half price is offset by 2-3x token usage, costing more than Sol — lordpuddingcup · 2026-10-27
- Fable 5.5 rumored to ship this week as partners still test — kimmonismus · 2026-10-27