Tencent's Hy4 Brings 1M Context and Agent Tools at a Tenth the Cost
Tencent's Hunyuan Hy4 scores 35 on Artificial Analysis Intelligence Index v5, ranking second among Chinese labs with a 1M context window and aggressive pricing.
20 stories
Tencent's Hunyuan Hy4 scores 35 on Artificial Analysis Intelligence Index v5, ranking second among Chinese labs with a 1M context window and aggressive pricing.
A new open-source agent skill breaks any short video into shots, flags what was filmed, AI-generated or coded, and hands back a remake plan.
Antigravity's IDE now renders .ipynb files directly in the file viewer, showing code, Markdown, tables, images, and outputs without external tools.
A new framework shows how to plug pretrained diffusion models into classical image reconstruction solvers, with injected noise helping iterates escape saddle points.
A trending community LoRA for Krea 2 brings consistent Japanese character generation to both the undistilled Raw base and the fast Turbo checkpoint.
Researchers at WeChat Vision, Tencent describe a 31.7B-parameter image foundation model that handles generation, editing, and multi-reference synthesis on one GPU.
Benji Taylor opened a public collection of Agent Skills for coding agents, led by Stageshow, a React toolkit for building faux product UI stills and cursor demos.
FAMDA plugs Segment Anything and Depth Anything into a self-training loop, producing tiny multi-task models that beat foundation models on real driving scenes.
Locals is an open-source local inference server for Apple Silicon built around multi-agent workloads, hitting roughly 3x the throughput of mlx-lm on Macs.
A new interpretability method trains a language model to generate text that provokes any target direction inside another model, verifiable by forward pass.
Nace.AI releases four open decision models that return calibrated probabilities over options in milliseconds, skipping autoregressive decoding entirely.
OpenAI's Images 2.6 adds native GIF generation, reusable reference images, and two API models that cut generation time by nearly half.
Engram's new research shows a short RL phase teaches small models to use their notes better, beating frontier models at 5-10x lower cost.
OpenAI's autonomous pipeline found 948 Linux kernel privilege escalation bugs, wrote exploits, and generated patches for $1.2M in API spend.
Together AI unveils Together Compiler, an agent-driven system that continuously retunes inference kernels as models, traffic, and GPUs change underneath them.
Xiaomi's new 0.8B vision-language model matches multi-billion parameter OCR systems on document parsing and visual question answering benchmarks.
Epoch AI rebuilt its Capabilities Index on a Bayesian model, delivering tighter uncertainty bands and head to head probabilities between frontier models.
Anthropic's biology lab used Claude to map the first 3D structure of a vault particle from bacteriophages, cutting months of work into days.
Cognition's new coding model runs 3x faster and 5x cheaper than SWE-2, beating all US open-weight models at its size.
Claude Code now uses Haiku 5.5 with short notes instead of full transcripts to refresh subagent status lines, cutting update costs by about 99%.