Interconnects: Open model roundup and reading list chart a licensing split and a 4-6 month open-closed gap

Interconnects surveys new open models (Motif-3, GLM-5.3, Hy4-preview and others) amid a licensing divergence—Western labs opening up, Chinese frontier labs restrictive—and publishes a reading list noting the open-closed frontier gap has narrowed to roughly…

Two Interconnects posts frame the state of open-weight AI as of September 2026. The first (Sept 8) surveys new open models: Motif-3 (MIT license, strong scores for its size), GLM-5.3 (switched from MIT to a custom license with a $10 billion revenue threshold and an undefined 'affiliates' clause requiring Z.AI security review), and Tencent's Hy4-preview (competent but prone to overthinking), plus dots3-note-prev from RedNote/Xiaohongshu (won IMO 2026 with a perfect score), Qwen3.8-Flash-Next (125B-A6B with GDN and Qwen Sparse Attention), NVIDIA Nemotron-3.5-Lightning-30B-A3B-BF16, and Ling-3.0-flash. Its core theme: Google and Meta adopted Apache 2.0 while Chinese frontier labs (Zhipu, Kimi K3, MiniMax M3) adopted restrictive commercial licenses. The second post (Sept 11) is a curated reading list on open-model strategy, licensing gradients, open-weight safety, adoption data, and Chinese open-source history. It notes leading open models have come from Chinese labs since roughly 2024 (citing Kimi K3 and GLM-5.2/5.3), that the open-closed gap has narrowed to roughly 4-6 months, and that Western adoption of Chinese models—Perplexity's use of DeepSeek R1 and Thomson Reuters moving to Qwen—has drawn lawmaker probes at DoorDash, Airbnb, Anysphere/Cursor, and Apple.

• GLM-5.3 switched from MIT to a custom license with a $10 billion revenue threshold and an undefined 'affiliates' clause requiring Z.AI security review
• Motif-3 ships under MIT license with strong scores for its size
• Tencent's Hy4-preview is competent but prone to overthinking
• dots3-note-prev from RedNote/Xiaohongshu won IMO 2026 with a perfect score
• Qwen3.8-Flash-Next previews a 125B-A6B architecture with GDN and Qwen Sparse Attention; other releases include NVIDIA Nemotron-3.5-Lightning-30B-A3B-BF16 and Ling-3.0-flash
• Google and Meta adopted Apache 2.0 licensing while Chinese frontier labs (Zhipu, Kimi K3, MiniMax M3) adopted restrictive commercial licenses
• The open-closed frontier gap has narrowed to roughly 4-6 months, per Interconnects' Sept 11, 2026 reading list
• Leading open models have come from Chinese labs since roughly 2024, including Kimi K3 and GLM-5.2/5.3