Track global AI model launches, company moves, open-source projects, and infrastructure updates worth paying attention to.
DeepInfer powers DeepSeek V4.1 Flash at scale. Its open release gives developers practical tools for coordinating accelerators, serving more users, and managing inference costs.
An uncensored DeepSeek V4.1 Flash weight release claims 320/320 HarmBench compliance, but its own MMLU score drops 3.86 points and testing requires four H200 GPUs.
Manus has introduced Cue alongside Manus 2.0, a personal AI Agent with its own phone number, email address, wallet, and computer. The product aims to move AI from one-off answers toward persistent, real-world assistance.
ZCode was accused of uploading workspace snapshots, Git history, and LFS data during codebase indexing. The official response says the issue is fixed, but public reverse-engineering evidence leaves important questions unanswered.
Ox Alpha has been identified as Z.ai's GLM-5.3-Flash. During its anonymous free preview, OpenCode logged 503K users, 13.12M completed sessions, 44T tokens, and a 10.6% share.
GLM-5.3 is here. Zhipu keeps the 744B-A40B base and scales up post-training instead, with a 1M context window, 128K output limit, 28.3 on Terminal-Bench 3.0, and 66.9 on DeepSWE v1.1.
FFmpeg 9.0 "Lei" is officially released with new LCEVC, Animated WebP, Vulkan, CUDA, AMF, and professional video capabilities. The codename honors Chinese multimedia developer Lei Xiaohua on the tenth anniversary of his death.
DeepSeek-V4-Flash reaches 82.7 on Terminal-Bench 2.1, beats GLM-5.2, approaches Claude Opus 4.8, and costs just ¥1 input / ¥2 output per million tokens for agents.
OpenAI cuts GPT-5.6 Terra by 20% and Luna by 80%. Luna now costs $0.20 input and $1.20 output per 1M tokens, while Sol pricing and subscription plans stay unchanged.
Kimi K3's full weights and 47-page technical report are out. We unpack its 2.8T MoE, 104B active parameters, 1M context, agent RL, deployment demands, and license limits.
