Five frontier-class open-weight LLMs shipped in last 30 days — Llama 4, Qwen 3.5, DeepSeek V4, Gemma 4, Mistral Medium 3.5
The last 30 days have seen five frontier-class open-weight LLMs ship: Meta Llama 4 (Scout and Maverick variants), Alibaba Qwen 3.5, DeepSeek V4 (Pro and Flash under MIT license), Google Gemma 4 (26B / 14GB optimized for accessibility), and Mistral Medium 3.5. The volume is unprecedented — the open-weight frontier tier now has more model families than the closed-weight Western frontier, and Apache 2.0 has won the permissive-license race.
The license consolidation around Apache 2.0 is the structural piece. Through 2024 the open-weight frontier was a patchwork of bespoke licenses with subtly different commercial-use restrictions, which slowed enterprise adoption. By mid-2026 the dominant licenses are Apache 2.0 (Qwen, Gemma, Mistral, Llama variants), MIT (DeepSeek V4), and a small remainder of restricted licenses with declining strategic relevance. For enterprises, the implication is that frontier open-weight models can now be evaluated on capability and ecosystem fit rather than on legal review — which collapses the procurement timeline from quarters to days.
The capability distribution across the five families is what makes the market actually competitive. Llama 4 (Scout, Maverick) leads on instruction-following for general-purpose deployments. Qwen 3.5 leads on coding and agent benchmarks. DeepSeek V4 leads on reasoning at the MIT-licensed frontier tier. Gemma 4 at 26B is the deployable-on-laptop choice for on-device inference. Mistral Medium 3.5 leads on European-language tasks and on commercial terms favorable to EU enterprise. The enterprise self-hosting market now has five strong choices where in 2024 it had effectively one (Llama 3.1 405B) — and the competitive dynamic between them drives the per-token economics down for everyone.
Hugging Face Blog — Best Open-Source LLM Models 2026 → · O-Mega — Top 10 Open Source LLMs 2026 DeepSeek Revolution → · Stormap AI — Game-Changing Open Source AI Models of 2026 →