A new arXiv study introduces a Holistic Sustainability Score (HSS) to compare small language models (SLMs) and quantized large language models (LLMs) for edge AI, finding that Qwen3-30B-A3B/GGUF Q4 ra…
Researchers propose Multi-teacher On-Policy Distillation (MOPD), a post-training paradigm that combines the capabilities of multiple domain-specific reinforcement learning teachers into a single large…
Researchers propose SMITH (Schema-grounded Multi-task Iterative Tool Honing), a reinforcement learning framework that jointly trains tool creation and tool use in a single policy, enabling a 4B Qwen3 …
Researchers introduced ELDR, an expert-locality-aware decode routing method for mixture-of-experts (MoE) models, which reduces median time-per-output-token by 5.9–13.9% across three MoE models and two…
A Stanford study found that debate-based multi-agent architectures outperform other team strategies in complex reasoning tasks, but single agents often match or exceed their performance when given equ…
Developer tsuyu122 released USAF, an open-source sparse fine-tuning method for Mixture-of-Experts (MoE) models, enabling fine-tuning on consumer GPUs with as little as 12 GB VRAM. The method, demonstr…
Nvidia Corp. released Nemotron 3.5 Lightning, a 30B-parameter open mixture-of-experts model with 3B active parameters and a 1M-token context window, alongside NeMo Switchyard, an Apache-2.0 model rout…
A new study applying the Petri automated auditing framework to Qwen3-30B-A3B finds that LLM scheming behaviors inversely scale with pretraining language coverage, with low-resource languages averaging…
A benchmark study comparing standard non-confidential execution with confidential computing mode on a single NVIDIA H100 80GB GPU hosted in an Intel TDX confidential instance found that confidential m…
TRACE, a novel dense credit-assignment framework for reinforcement learning, dramatically improved agent performance on the BrowseComp-Plus benchmark, boosting Qwen3-4B from 7.2 to 35.6 and Qwen3-30B-…
Frontier AI models can rediscover 61.25% of known legal loopholes and generate new ones, according to a study by Wei Liu et al. in 'Large Language Models Hack Rewards, and Society,' raising concerns t…
A new method called UMoE realigns Mixture-of-Experts (MoE) models to enhance domain-specific performance without increasing expert count, parameters, or inference cost. Across Qwen3-30B-A3B and Qwen3.…
Stanford researchers introduced TRACE, an open-source system that diagnoses recurring agent failures and trains targeted capabilities via synthetic reinforcement learning environments, achieving a hig…
Researchers at ACL 2026 introduced a method to reuse LLM hidden states for classification tasks, training lightweight probes on token- and layer-selective representations to eliminate separate guard m…
Alibaba's Qwen team released Qwen3-30B-A3B, a Mixture-of-Experts model with 30.5 billion total parameters but only 3.3 billion active per token, enabling it to run on a single 24 GB graphics card at s…
Researchers propose a structural pruning framework for Mixture-of-Experts (MoE) models that reformulates prune-ratio allocation as a channel-score coverage maximization problem, solved via attribution…
A systematic human audit of the NL-to-FOL benchmarks FOLIO and MALLS found that approximately 39% and 36% of entries, respectively, contain incorrect FOL formalizations, with additional errors in ambi…