Local AI inference at 32B-parameter quality, no cloud API required: University of Waterloo researchers released PAW on July 2 ...

Fable 5's chain of thought has leaked, showing math-like shorthand, while its three-layer defense classifiers block most jailbreak attempts.

DSpark can make decoding faster, but acceptance quality still determines how much speed the system actually realizes.

MiniMax officially released MiniMax M3 on June 1, 2026. The model introduces MSA (MiniMax Sparse Attention), a new sparse attention architecture that gives M3 a 1M-token context window. M3 also ...

Speculative decoding is a technique for speeding up large language model inference. A small, fast draft model proposes several tokens. The large target model verifies them in parallel. If accepted, ...

SU-01 is a 30B-A3B olympiad reasoning model trained with a simple and unified post-training recipe for mathematical and scientific problem solving. The goal is to turn a broadly capable post-trained ...

Google released Multi-Token Prediction (MTP) drafters for Gemma 4, delivering up to a 3x speedup at inference without any degradation in output quality. The technique—called speculative decoding—uses ...

Mistral has released Mistral Medium 3.5, a dense 128B parameter model that pulls instruction-following, reasoning, coding and vision into one set of weights. It replaces Mistral Medium 3.1 and ...

Last month, a founder I mentor sent me a screenshot of his OpenAI billing dashboard. The number was $2,847. For a single month. For a two-person startup. His product was barely in beta! I have been ...

You don’t notice good video compression—until it’s not there. For years, people have streamed high-resolution video without thinking about the tech behind it. But when companies clash over which ...