Tencent releases Hy3, a 295B-parameter model that it says is competitive with GLM-5.1 and 5.2, under the Apache 2.0 license, following a preview launch in April
THE SO WHAT
A 295B-parameter, Apache 2.0–licensed Hy3 that benchmarks against GLM-5.x pushes truly open-weight, high-end models into enterprise-safe territory. If you’re in EMEA or Korea and were blocked by prior Chinese licenses, your open model shortlist just changed — legal and infra teams should revisit their default stack.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIRunway Expands Global AI Footprint
Runway putting nearly $300M behind hubs in London, Tokyo, and Paris is a bet that high-end video gen will be a global creative infrastructure layer, not a niche tool. If you’re in media, marketing, or design, assume local talent and brand teams will have first-class access to these capabilities and plan workflows around that, not SF-only pilots.
Applied AISpaceX AI device prototype: What we know so far
If SpaceX is exploring a mobile AI device, that’s another large-scale distribution channel experimenting with vertically integrated hardware, connectivity, and models. For consumer product builders, the risk is fragmentation—assume more proprietary AI surfaces beyond iOS and Android and design your roadmap for selective, not universal, presence.
Applied AIIf you use Google, you’re training its AI. Here’s how to opt out.
Google quietly widening the data it can use for AI training — with an opt-out, not opt-in — formalizes what many users assumed but hadn’t confronted. Product teams relying on third-party data should expect more of these consent toggles and design data strategies that survive a world where a meaningful share of users say no.
Cybersecurity firm says it found 'the first documented case' of AI agentic ransomware
Agentic ransomware is the nightmare scenario security teams have been modeling — malware that can adapt, plan, and iterate without human operators. If you own critical infrastructure or high-value data, assume dwell time and attack creativity just went up and revisit your assumptions about what “automated” defense needs to look like.