Parameter– tag –
-
Thinking Machines Lab Unveils Inkling, a New Open-Weight 975B MoE LLM, Achieving Superior Benchmark Performance Over GLM-5.2
Sebastian Raschka Germany Overview Thinking Machines Lab has announced Inkling, a new open-weight Large Language Model (LLM) with approximately one trillion parameters. This 975B Mixture-of-Experts (MoE) model demonstrates solid performa... -
July 2026 AI Trends: Shift from Large Models to Practicality, Cost-Efficiency, Multimodal AI, and Autonomous Agents Emerge
ZoneTechify Blog USA Overview July 2026 AI trends show a significant pivot from larger models to practical, cost-efficient, and reliable AI solutions, with inference costs plummeting. Multimodal models, capable of processing text, image,... -
arXiv Paper: Looped World Models (LoopWM) Achieve 100x Parameter Efficiency, Resolving Fidelity-Cost Tension in Long-Horizon AI Simulation
arXiv (cs.LG) International Overview Researchers have introduced Looped World Models (LoopWM), the first looped architectures for world modeling that resolve the tension between faithful long-horizon simulation and deployment costs. Loop... -
arXiv Preprint ‘ThousandWorlds’ Introduces New Benchmark for AI Climate Emulation of Potentially Habitable Exoplanets
Astrobiology International Overview An arXiv preprint introduces "ThousandWorlds," a new benchmark for evaluating machine learning models designed for climate emulation of potentially habitable exoplanets. This initiative provides a stan... -
“Satisfiable Drift” Problem Emerges in Multi-Turn Reasoning of LLMs, Revealed by Novel DRIFT-Bench Benchmark
AI Accelerator Institute International Overview Researchers developed DRIFT-Bench, a solver-instrumented benchmark with 816 test problems across three constraint domains, to evaluate multi-turn reasoning in open-weight models ranging fro... -
Zyphra Unveils ZAYA1-8B: Small AI Model Achieves Large-Scale Performance with 700M Active Parameters
ライブドアニュース (GIGAZINEからの転載) Japan Overview US AI startup Zyphra has released ZAYA1-8B, a compact inference language model trained on AMD GPU infrastructure. Despite being an 8-billion-parameter Mixture of Experts (MoE) model,...
12