Google introduces Gemini 4 Hydrogen, a model designed for rapid coding, debugging, and multimodal workflows, available via Google AI Studio and Antigravity.

This article was created by AIHubMix. The figures are from the Artificial Analysis snapshot dated November 11, ...

OverviewA new AI model “Claude Fable 5.2” has been announced, and it is making waves for achieving a high score of 65 points ...

NVIDIA Dory 4B uses a recurrent looping mechanism to achieve a 61.8% SWE bench score, but decoding speed drops from 188 to 78 tokens per second.

Researchers have introduced NVCR, an extensible C++20 native runtime that accelerates neural video compression execution ...

A Foundry on Windows engineering report explains how mixed-precision quantization lets MAI-Code-1.1-Flash run on RTX Spark hardware.

Ricursive's Ripple-65B Preview ships 129.6 GB of Apache 2.0 weights claiming 1M context, but the first hosted API stops at 16 ...

NVIDIA shows 3 ways to speed up MoE inference on DGX Spark, lifting Ling-3.0-flash GSM8K decode throughput 42.5%.

Mistral AI on October 29, 2026 released the weights of Mistral Large 4, its flagship 1-trillion-parameter natively multimodal model known as Le Chonk, making the checkpoint downloadable from Hugging ...

DeepSeek open-sourced DeepInfer, its Apache-2.0 inference framework for V4.1-Flash, with support for NVIDIA SM100 and Huawei Ascend 950 chips.

DeepSeek-V4.1-Flash-20261017 hits 1670 points on Code Arena WebDev, ranking #1 open model and passing Kimi K3 (Max) at about 10x lower cost.