Import AI 465: Open vs closed gaps; Kimi K3; Demis' big policy plan

UK government analysis shows open-weight AI models (GLM-5.2, DeepSeek V4-Pro) are closing the cybersecurity capability gap with proprietary frontier models—from 6-10 months behind in 2025 to 4-7 months behind in 2026. Chinese model Kimi K3 demonstrates frontier-level performance across benchmarks, matching or approaching Claude and GPT equivalents.

This Wire brief sits within Fusion42's coverage of AI Frontier Models. Wire is Fusion42's founder-focused intelligence feed: each story is connected to the funds and startups it names — every one with a live profile on Raise or Scout — so founders can follow the capital and the momentum behind the headline rather than just the headline itself. Wire analysis is one of the live surfaces Arthur reasons over.