Engineering
LFM2.5-VL-3B for Better and Faster Vision Capabilities for the Edge
Liquid AI introduced LFM2.5-VL-3B, a vision-language model that pairs a SigLIP2 400M NaFlex vision encoder with the LFM2.5-2.6B text backbone.…
Engineering
A One-Line Change Moved a Drug-Discovery Benchmark by 0.21 AUROC
FINAL-Bench released LEADBOARD, a drug-property-prediction benchmark spanning 21 boards and more than 18,000 held-out test compounds. The team found that…
Engineering
DeepMind Partners With Fenris Creations to Advance AI Research Inside EVE Online
Google DeepMind announced a research partnership with Fenris Creations, the studio behind EVE Online, to use the game’s persistent virtual…
Engineering
EnvHarness: A Programmable Layer for Reshaping Agent Training Environments
Researchers at Google introduce EnvHarness to address the problem that agent learning environments are static and don’t adapt to an…
Engineering
SWE-bench Science Shows Coding Agents Still Struggle to Repair Scientific Software
Researchers introduced SWE-bench Science, a repository-level benchmark comprising 119 tasks drawn from 98 GitHub repositories across 20 scientific domains, to…
Engineering
MemTrapBench Finds That Giving LLM Agents Memory Can Backfire
Researchers introduce MemTrapBench, a benchmark showing that retrieved memories can actually harm large language model task performance even when the…
Engineering
SkillEvo Lets AI Agents Learn From Multi-Turn Interaction Failures
Tencent researchers present SkillEvo, addressing the problem that AI agent skills typically fail to improve from interaction failures beyond a…
Engineering
Repo0 Generates Entire Software Repositories From Natural Language Requirements
Researchers at Shanghai Jiao Tong University present Repo0, a system for generating complete software projects with proper modular architecture directly…
Engineering
GLM-5.3 Finds 2,436 Real Security Bugs, Some Dormant for Decades
Z.ai’s GLM-5.3 model was trained with post-training scaling improvements aimed at vulnerability detection, achieving 84.5% on the CyberGym benchmark for…
Engineering
MemGraphRAG Uses a Society of AI Agents to Build Cleaner Knowledge Graphs
Researchers address a weakness in graph-based retrieval-augmented generation systems, where automatically constructed knowledge graphs often suffer from thematic irrelevance, logical…
Engineering
LFM2.5 Q4_0 Checkpoints from Quantization-Aware Distillation
Liquid AI released Q4_0 GGUF checkpoints for four LFM2.5 models using Quantization-Aware Distillation (QAD), a technique that distills a high-precision…
Engineering
Same Cluster, 33 Points More Utilization: What Changed Was the Order
Dharma AI describes a constraint-aware GPU allocation system that improved cluster utilization by up to 33 percentage points compared to…