/2 VibeThinker-3B is a 3-billion-parameter model proving that top-tier verifiable reasoning can be efficiently compressed into small language models. It uses a "Spectrum-to-Signal" pipeline with supervised fine-tuning, reinforcement learning, and self-distillation. By optimizing
RESEARCH
-
LoopWM: First looped transformer for stable long-horizon world modeling
By
–
/1 Meet LoopWM: The first looped transformer for stable, long-horizon world modeling! Current world models struggle with deep computation costs and compounding errors. Enter Looped World Models (LoopWM), which iteratively refine latent states through a parameter-shared
-
Sakana Fugu: Single Model API for Multi-Agent Systems
By
–
Sakana Fugu: Providing Multi-Agent Systems as a Single Model API https://
sakana.ai/fugu-release/#
Japanese
… Related article: Sakana Fugu Technical Report https://
github.com/SakanaAI/fugu/
blob/main/Fugu_technical_report.pdf
… -
New AI math tool sharpens image editing, drug discovery, simulations
By
–
New #AI math tool could sharpen image editing, drug discovery and simulations
by @ClarksonUniv @TechXplore_com Learn more: https://
bit.ly/4gdVImD #ArtificialIntelligence #MachineLearning #ML -
A handful of French people understood AI 10 years ago
By
–
10 years ago, a handful of French people understood the AI revolution but no one was listening @mikiane @StephaneMallard @OlivierBabeau @nbouzou @RobinRivaton @raphaeldoan @ThomasScialom @babgi @oliviersichel
-
NVIDIA PiD decodes AI latents to 4K in under a second
By
–
Can you decode AI latents to 4K images in under a second? NVIDIA researchers introduce PiD, a pixel diffusion decoder that unifies decoding and upscaling into one fast step. It converts 512×512 latents to 2048×2048 pixels in under 1 sec on a consumer RTX 5090 — 6x faster than
-
Google Crushed by Open-Source Chinese Models Despite Infinite Resources
By
–
Google: Infinite data. Unlimited compute. Still crushed by open-source Chinese models
-
@testingcatalog — 2026-06-22
By
–
URGENT : Sakana AI a annoncé les systèmes Sakana Fugu et Sakana Fugu Ultra, qui obtiennent des performances équivalentes à Claude Fable 5 et Mythos 5 sur de nombreux benchmarks. > Sakana AI est un laboratoire d’IA japonais, et Fugu est un modèle d’orchestration entraîné pour
-
Benchmark for testing agent self-evolution
By
–
Allow me to present our Benchmark, which is specifically designed to test the self-evolution capability of Agents, which is absolutely crucial. This year, we have constantly explored the direction of self-improvement / self-evolution, because we have
-
Sakana AI’s Fugu Ultra ties with Fable 5
By
–
Sakana's Multi-Agent ties with Fable 5: Sakana AI's Fugu Ultra may not be a new state-of-the-art model in the classic sense. It is more like a learned orchestration layer that transforms multiple state-of-the-art models into a coordinated team of agents.