NVIDIA has published the training and inference recipes behind Nemotron 3 specialist systems that it reports scored above gold thresholds at both IOI 2026 and IMO 2026, combining fine-tuning with iterative generation, ve…
OpenAI published 722 mathematical manuscripts grouped into 372 result families from an unreleased internal frontier model, alongside Lean formalizations, reasoning summaries and production details, while explicitly warni…
Cohere Labs reports that Tiny Aya L2-Thinker, a 3.35B-parameter research model, reasons in the prompt language more than 93% of the time across six benchmarks and 60 languages, using a training mix that combines English …
