THE 60-SECOND VERSION
● Tencent Hy4 Preview: 770B total, 49B active, 1M context, Apache 2.0.
● Google to European users: the largest reduction in quality of service in Search's 29-year history.
● Mistral raised $3.5 billion. China announced a plan to quadruple national compute.
● 6 days to Claude Code weekly limits settling lower.
Tencent ships a flagship under Apache 2.0
Hy4 Preview is an open-weight mixture-of-experts model at 770 billion total parameters with 49 billion active and a 1 million token context window, released under Apache 2.0.
That licence is the notable part. Four earlier open-weight flagships this summer used custom, modified or unstated terms — Qwen 3.8-Max custom, Kimi K3 modified MIT, GLM-5.3 flagship still unstated. Apache is standard, unmodified and familiar to legal teams, which lowers the cost of building on it more than a few benchmark points would.
Where this leaves the open-weight field →
Google tells 450 million Europeans Search got worse
Google rolled out revamped EU search results to comply with the Digital Markets Act and a €460 million self-preferencing fine issued in July, telling Reuters the changes represent the largest reduction in quality of service in Search's 29-year history.
That is a remarkable thing for a company to say about its own product. Whether it is an honest assessment or a lobbying position depends on your reading of the incentives, and both interpretations are available.
Also today
- Mistral raised $3.5 billion to stay competitive in the frontier race.
- China announced a plan to quadruple national compute capacity.
- Arm unveiled the Neoverse CSS N4, codenamed Ranger, packing 128 cores on TSMC N3P for cloud, networking and agentic-AI workloads.
- OpenBMB released MiniCPM5-2B, a 2.5B on-device model averaging 53.9 across 34 benchmarks and topping the sub-4B open leaderboard.
- GLM-5.3 went live in Perplexity Computer for long-context multimodal agent workloads.
- AI memory demand is raising consumer electronics prices, as manufacturers redirect capacity from conventional DRAM toward high-bandwidth memory for data centres.
The week ahead
| Date | What happens |
|---|
| Sept 14 | Claude Code weekly limits settle 17 percent below current levels |
| Sept, unconfirmed | OpenAI listing window. No public S-1 on EDGAR |
| Oct 1 | OpenAI vs Apple hearing |
| Nov 12 | OpenAI models leave Cursor |
FAQ
What is Tencent Hy4 Preview?
An open-weight mixture-of-experts model at 770 billion total parameters with 49 billion active and a 1 million token context window, under Apache 2.0.
Why did Google say Search got worse?
It rolled out revamped EU results to comply with the Digital Markets Act and a €460 million fine, and described the change to Reuters as the largest reduction in quality of service in Search's 29-year history.
What is the next confirmed AI deadline?
14 September, when Claude Code weekly limits settle at 25 percent above the pre-May baseline — 17 percent below current levels.