Excited to see the image reasoning in o3 and o4-mini!!🤩 We introduced this idea a year ago in visual Sketchpad ( https://t.co/v5TYvzyjGM). Excited to see @OpenAI baking this into their model through agentic RL. Great work! And yes, reasoning should be multimodal! Huge shoutout

Today we’re excited to unveil a new generation of Segment Anything Models: 1️⃣ SAM 3 enables detecting, segmenting and tracking of objects across images and videos, now with short text phrases and exemplar prompts. 🔗 Learn more about SAM 3: https://t.co/tIwymSSD89 2️⃣ SAM 3D

Introducing SAM 3D, the newest addition to the SAM collection, bringing common sense 3D understanding of everyday images. SAM 3D includes two models: 🛋️ SAM 3D Objects for object and scene reconstruction 🧑‍🤝‍🧑 SAM 3D Body for human pose and shape estimation Both models achieve

COLM Keynotes: Luke Zettlemoyer Mixed-modal Language Modeling https://t.co/8FdhhrfOnG

🚀Our Code and Weight is public available. We provide detailed examples, and try out at https://t.co/0jE0CrSTLY

We are hiring a PhD research intern (summer 2026) at Meta FAIR to work on frontier multimodal generation models! Apply here: https://t.co/M5htDrYUBR Feel free to DM me if you have any questions!

We’re hiring a Research Intern (Summer 2026) at FAIR to advance the frontiers of Multimodal Generative AI. Learn more and apply here: https://t.co/aVRV04lp6V If you have questions, feel free to DM me or meet me at the RAM 2: Reasoning, Attention & Memory workshop at COLM.

Our team at Meta FAIR is hiring a PhD research intern for 2026. The topics broadly involve multimodal generative AI (e.g., video/image generation in addition to text), with flexible approaches across architecture/data/algorithms. Please apply via the link below, and feel free to

Transfusion combines autoregressive with diffusion to train a single transformer, but what if we combine Flow with Flow? 🤔 🌊OneFlow🌊 the first non-autoregressive model to generate text and images concurrently using a single transformer—unifying Edit Flow (text) with Flow

As computer-use agents (CUAs) handle critical digital tasks, open research is key to study their capabilities, risks. 🚀After a year, we release OpenCUA: 1) largest CUA dataset/tool, 2) training recipe, 3) ~SOTA model on OSWorld. Released to drive transparent,safe CUA research!

Introducing DINOv3: a state-of-the-art computer vision model trained with self-supervised learning (SSL) that produces powerful, high-resolution image features. For the first time, a single frozen vision backbone outperforms specialized solutions on multiple long-standing dense

Introducing Genie 3, the most advanced world simulator ever created, enabled by numerous research breakthroughs. 🤯 Featuring high fidelity visuals, 20-24 fps, prompting on the go, world memory, and more.

🎉 We’re excited to introduce BLAB: Brutally Long Audio Bench, the first benchmark for evaluating long-form reasoning in audio LMs across 8 challenging tasks, using 833+ hours of Creative Commons audio. (avg length: 51 minutes).

Compression is the heart of intelligence From Occam to Kolmogorov—shorter programs=smarter representations Meet KARL: Kolmogorov-Approximating Representation Learning. Given an image, token budget T & target quality 𝜖 —KARL finds the smallest t≤T to reconstruct it within 𝜖🧵

Having trouble dealing with the excessive token number when processing a video? Check out our paper that is accepted by ICCV 2025 with an average score of 5.5! We tokenize video with tokens grounded in trajectories of all objects rather than fix-sized patches. Trained with a

When cooking multimodal models, one big headache I found is that the evaluation benchmarks are not reliable — especially for tasks like interleaved generation.😢 The authors make big effort on reliability, crafting one eval pipeline for each task. 🔥For example, they even

We introduce MMMG: a Comprehensive and Reliable Evaluation Suite for Multitask Multimodal Generation ✅ Reliable: 94.3% agreement with human judgment ✅ Comprehensive: 4 modality combination × 49 tasks × 937 instructions 🔍Results and Takeaways: > GPT-Image-1 from @OpenAI

The most shocking figure: how can an LM calculate 3sqrt(13) correctly up to 15 decimal places? Unless something in pretraining is very similar ... The paper is not questioning RLVR - correct reward works for all LMs. It just questions studies that experimented ONLY with Qwen.