News
Synthetic Sciences Releases Open Science: An Open-Source, Model-Agnostic AI Workbench for Machine Learning, Biology, Physics, and Chemistry Research
2+ hour, 43+ min ago (435+ words) Synthetic Sciences has released Open Science, an open-source AI workbench for scientific research. It is licensed under Apache 2. 0 and runs on your own infrastructure. The research team frames it as an open alternative to Anthropic's Claude Science, launched in late…...
Meituan Releases Long Cat-2. 0: A 1. 6 T-Parameter Open Mo E Model with Native 1 M Context and Long Cat Sparse Attention
10+ hour, 25+ min ago (383+ words) Meituan has released Long Cat-2. 0, a large-scale Mixture-of-Experts (Mo E) language model. It carries 1. 6 trillion total parameters and activates about 48 billion per token. The model targets agentic coding: code understanding, generation, and execution inside agent workflows. Two facts stand out. First,…...
Google AI Introduces Tab FM: A Hybrid-Attention Tabular Foundation Model for Zero-Shot Classification and Regression
5+ day, 1+ min ago (511+ words) Google Research introduced Tab FM, a foundation model built for tabular data. Tab FM performs classification and regression without dataset-specific training. Every prediction comes from a single forward pass. The model reframes tabular prediction as an in-context learning problem. It…...
Meta AI Releases Brain2 Qwerty v2: A Non-Invasive MEG Brain-to-Text Pipeline Decoding Typed Sentences at 61% Word Accuracy
5+ day, 23+ hour ago (494+ words) Meta's FAIR lab releases an end-to-end deep learning system that reads typed sentences from magnetoencephalography signals, no surgery required. Brain2 Qwerty v2 is a brain-to-text decoder. It maps raw brain activity to characters, then to words and sentences. Meta trained it on…...
Liquid AI Ships LFM2. 5-230 M with llama. cpp, MLX, v LLM, SGLang, and ONNX Support for On-Device Inference
1+ week, 1+ day ago (573+ words) Liquid AI's smallest model yet runs at 213 tokens per second on a Galaxy S25 Ultra and 42 on a Raspberry Pi 5. Liquid AI shipped LFM2. 5-230 M, it's the company's smallest model to date. The release targets a specific job: running agentic tasks on…...
Deep Seek Releases DSpark, a Speculative Decoding Framework That Accelerates Deep Seek-V4 Per-User Generation 6085% Over MTP-1
1+ week, 1+ day ago (964+ words) Deep Seek released DSpark, a speculative decoding framework, with open-source checkpoints and training code. It is a serving optimization, not a new model. The checkpoints Deep Seek-V4-Pro-DSpark and Deep Seek-V4-Flash-DSpark reuse the existing V4 weights, with a draft module…...
Baidu Releases Unlimited OCR, a 3 B Model That Keeps the KV Cache Flat for Long-Document Parsing
1+ week, 4+ day ago (598+ words) Baidu open-sourced Unlimited OCR on June 22, 2026. It reads long documents in a single pass by holding the attention cache flat. Most end-to-end OCR models slow down as output grows. Each generated token adds to the KV cache. Memory rises and…...
DFlash Speculative Decoding Drafts Whole Token Blocks in Parallel for Up to 15x Higher Throughput on NVIDIA Blackwell
1+ week, 5+ day ago (756+ words) Autoregressive large language models generate text one token at a time. Each token waits for the one before it. This serial loop leaves modern GPUs underused and keeps inference slow. The cost grows worse with long Chain-of-Thought reasoning models. Their…...
Prime Intellect Releases prime-rl 0. 6. 0 to Train Trillion-Parameter Mo E Models on Agentic RL Workloads
1+ week, 6+ day ago (550+ words) The research team trained GLM-5 on SWE tasks at up to 131k sequence length. Step times stayed under five minutes. The batch size was 256 rollouts. The run used only 28 H200 nodes. prime-rl is an open framework for asynchronous reinforcement learning. It post-trains…...
Moon Math AI Open-Sources a HIP Attention Kernel for AMD MI300 X That Beats AITER v3 on Every Shape and Rounding Mode
2+ week, 35+ min ago (684+ words) Moon Math AI team has released a bf16 forward attention kernel for AMD's MI300 X GPU. It is written in HIP, not hand-written assembly. The code is open-source under the MIT license. The Moon Math. ai team reports it beats AITER v3, AMD's…...