MBZUAI’s IFM releases open 29B PAN 2 world model
Apache 2.0 weights and code ship; IFM reports 34% full-plan completion, ahead of MiniMax-H3 on multi-step tasks.
1
Stories per day · 30 days
4 RSS
Relatedrobotics7gemini5google-deepmind4genie3genie-43video-generation3autonomous-driving2diffusion2libero2project-genie2simulation23d13d-world-models142ai1adversarial-attack1adversarial-robustness1agent-safety1behavioral-ai1
59 stories, newest first · one card per story
Stories tagged #world-model
PrimeSim world model simulator from large-scale cross-embodiment demonstrations
PrimeSim is an action-conditioned multiview world model simulator that runs at 10.1 FPS and transfers to held-out platforms after 100 to 200 gradient steps.
42AI comes out of stealth with $62.5M seed funding for behavioral world model
The Israeli startup 42AI has raised $62.5 million to develop a transformer-based world model that predicts human decision probabilities based on psychological tendencies.
1
RubricWorld improves LLM grading accuracy with world model-guided rubric optimization
Raises gpt-5-mini accuracy from 18.20% to 46.70% on ASAP2.0 exceeding baseline by 22.60 points.
Imperceptible goal-image attacks spoof latent state in LeWorldModel
Epsilon 4/255 perturbations spoof block pose in 73-95% of scenes and drop LeWorldModel planning success from 87% to 32%.
Tiny world model runs games on a Raspberry Pi Pico 2 W
The Hacker News post links to the tinyworldformer GitHub repo and had 42 points and 4 comments.
Google DeepMind introduces Genie 4 real-time world model
The title says Genie 4 includes native audio, but the captured page is only a cookie notice.
1
Google DeepMind unveils Genie 4 real-time world model as research preview
Only select testers have it now; Google AI Ultra subscribers and early Gemini API partners are due next, with SynthID on outputs.
1
Google DeepMind releases Genie 4 world model in research preview
Trusted testers get it now; Google AI Ultra and Gemini API partners follow in the coming weeks.
1
DeepMind announces Genie 4 real-time world model with native audio
A research preview goes to trusted testers first, then Google AI Ultra subscribers in Project Genie and early Gemini API partners.
1
RynnWorld-Latent conditions a world model on universal latent actions
Pretraining spans 23 datasets and about 553.5 million frames, then adapts to target robots without a shared action space.
MLE-World predicts ML experiment outputs without running training
Finetuned from Qwen3.5-4B, it matched held-out terminal output on 77.9% of programs versus 41.1% for larger prompted models.
Hy-Nomas generates multi-turn interactive video worlds from text and camera
The 8B diffusion model uses action-aware causal attention and a prompt enhancer; authors release weights, code, and benchmarks.
Video-Point Model forecasts multi-view video and 3D point trajectories
Point supervision on VPM-Data and VP-WorldBench improves video fidelity, 3D motion, and robot manipulation success.
OverSeer adds fixed auxiliary views to multi-agent world models
A delta-VAE encodes only changes from a static reference, cutting tokens for auxiliary views.
Dexterity from Supervised Play
DSP uses supervised play to collect robot data, plan with a world model, and match RL with far less interaction.
Can World and Action Modeling Benefit Each Other under Unified Causal Training?
OneWM shows unified causal training lets world modeling and action learning improve each other.
GeoState separates scene geometry from camera motion in 3D world models
Predicted states can be decoded under predicted or specified cameras without per-frame scale alignment.
SafeDriveVLA: Navigation-Conditioned World Model Dreaming for Conflict-Aware End-to-End Autonomous Driving
SafeDriveVLA uses world-model dreaming to refuse unsafe driving instructions, cutting collisions from 60 to 38.
WorldCast: Exploring Intelligent Visual Generation in World State Prediction
WorldCast ranks 12 models on visual world-state prediction, with GPT-Image-2.5 scoring 72.8.
BELIEFWEAVER improves long-horizon agent recovery from earlier failures
A distilled 9B student gained 9.8 points over the native baseline and cut estimated teacher API cost by 54.8%.
PINT: A Unified Model with Shared Predictive Intent for Robot Manipulation
PINT coordinates vision-language, video, and action models via shared Predictive Intent for stronger robot manipulation.
Odyssey-3 is a new generative world model that you can try for free
Odyssey opened a free preview of Odyssey-3, a 14-billion-parameter world model for real-time interactive environments.
WOVEN: Weaving Visual World Modeling into Multimodal LLMs
WOVEN trains multimodal LLMs on visual transitions, lifting 22 of 26 external benchmarks by up to 27.3 points.