Permissive flagship open-model licenses fell from 75% to 35% in months
bookwormengr · x · 2026-10-28
Analysis by 0xkydo shows frontier open-source licenses tightening fast: in February 75% of tracked labs had a permissively licensed flagship, now only 35% do. Since late July, six licenses added revenue caps targeting Model-as-a-Service operators only (Kimi K3 / Muse Spark Open / Step-5 at $20M, Qwen3.8-Max at $50M, GLM-5.3 at $10B, Qwen 3.8 Flash needing permission at any revenue). Individual users are mostly unaffected. The linked Open Model License Dashboard traces each lab's arc: DeepSeek went MIT from R1; Zhipu's GLM-5.3 requires security review for MaaS above $10B; MiniMax cycled through 7 licenses in one year; Alibaba's Qwen3.8 added MaaS/assistant triggers; Tencent's Hy3 returned to Apache 2.0; NVIDIA progressively loosened. Takeaway: frontier open source has retreated behind an open-weights-plus-commercial-license wall, while US open models are getting more permissive.
More from Models
- Creator spends $25,000 benchmarking reasoning_effort across models, open-sources trajectories — zainhas · 2026-10-29
- HF maintainer: EmbeddingGemma2 is actually a bidirectional continuation of Gemma 4 — tomaarsen · 2026-10-29
- Researcher: benchmark biases hide LLM flaws, like extreme-value numeric distractors — antoine_chaffin · 2026-10-29
- Sampura Research debuts safety-judge leaderboard built on 3957 tasks and 35613 human judgments — snikolov · 2026-10-29
- OpenAI engineer: GPT-5 felt slow not due to latency but because it emitted too many tokens — npew · 2026-10-29
- Leaker mark_k teases: expect something nice from OpenAI today — mark_k · 2026-10-29