ByteDance is training a mega AI model targeting roughly 10 trillion parameters, aiming to match top-tier frontier scale and compete head-on with Anthropic, though the project remains in early stages with final parameter count not yet locked in.
Moonshot AI closed a $3.5 billion round at a $34.9 billion valuation, accelerating IPO preparations. The raise is fueled by Kimi K3, a 2.8-trillion-parameter model billed as the world's largest open-source LLM at release on July 16.
Alibaba plans to keep its next Qwen model open-weight but charge the largest commercial users through a revenue-sharing arrangement, a notable shift in how Chinese labs monetize open-source AI. Qwen3.8-Max is expected to release next week under Apache 2.0.
Prime Intellect released Prime Agent, an open-source self-improving coding harness that scored 95.5% on the ARC-AGI-3 benchmark, outperforming human experts. The startup has $150M in total funding backing the project.
A SaferAI evaluation finds Z.ai's open-weight GLM-5.2 approaching GPT-5.5 and Claude Opus 4.7 on cyber and bio capability benchmarks, while carrying fewer safety mitigations than its closed-weight competitors.
The White House's still-unreleased AI oversight framework will only cover "state-of-the-art" models deemed national security risks, explicitly exempting lower-cost nonproprietary open models from heavy regulation, a win for the open-source AI community.
AMD acquired Taalas, a chip startup that hardwires AI model weights into silicon for inference, a method the company claims is 100x less expensive than training a frontier model. The move positions AMD to negotiate custom silicon deals with OpenAI, Anthropic, and Meta.