New AI tsunami: IBM, Alibaba, and Tencent drop massive models
Another week in tech, another tidal wave of neural networks hitting servers before engineers can even finish reading the documentation of the previous batch.
Thomson Reuters sank $40 million into Thomson, a dedicated legal and tax LLM built to keep enterprise systems from hallucinating statutory codes. Meanwhile, IBM launched Granite 4.2, cramming native reasoning into compact 3B to 30B parameter packages so corporate servers don't burst into flames under multi-step agentic workloads.
The open-weight heavyweight clash escalated just as quickly. Alibaba dropped the weights for Qwen3.8-Flash-Next as an architectural preview, tweaking attention and embeddings to cut inference bills. Not to be outdone, Tencent unloaded Hy4 preview, a staggering 770-billion-parameter MoE model with a 1-million-token context window designed for heavy office grinds, while Z.ai pushed both GLM-5.3 and its lightweight GLM-5.3-Flash.
Video generation took another sharp turn toward practical workflows. Alibaba rolled out Wan 3.0, capable of generating and editing continuous scenes up to 30 seconds, while Nvidia teamed up with the University of Manchester to release VideoNeuMat, extracting isolated 3D neural materials directly from video streams.
Voice and speech synthesis shrank down to microchips. Google launched Gemini 3.5 Transcribe to automatically strip filler words and stutters in 85 languages, while the 120-million-parameter Sopro V2 Turbo managed to clone voices in 300 milliseconds on a standard laptop CPU.
The relentless flood of hyper-specialized models proves that the dream of one single omnipotent chatbot running the entire world is dead, replaced by hundreds of miniature algorithmic cogs devouring enterprise workflows.
Source: GitHub
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.