Open-source LLM momentum grows
Meta’s Llama series and a new coalition between Nvidia and Mistral are lowering the bar to build with open LLMs, and firms like Cloudflare are publicly betting on open models — that’s a clear signal that Big Tech is funding and shipping open alternatives to closed APIs. The move reshapes both developer access and how startups build AI features without vendor lock-in. (usine-digitale.fr)(startuphub.ai)
NVIDIA unveiled the Nemotron Coalition at GTC on March 16, 2026, a multi-lab initiative that names Mistral AI, Perplexity, LangChain, Cursor, Reflection AI, Sarvam, Thinking Machines Lab and Black Forest Labs as inaugural members. (investor.nvidia.com)) The coalition’s first public project is a base model co-developed with Mistral, slated to be trained on NVIDIA DGX Cloud and released as an open-weight foundation underpinning the upcoming Nemotron 4 family. (investor.nvidia.com)) Mistral announced its own March 16, 2026 product push at NVIDIA’s GTC — including a new Forge enterprise training platform and an open-model roadmap that aligns with the Nemotron collaboration. (techcrunch.com)) Mistral’s ability to partner at this scale follows a €1.7 billion Series C in September 2025 that valued the company at €11.7 billion and included strategic investors such as ASML and NVIDIA. (mistral.ai)) Cloudflare announced on March 19, 2026 that Workers AI can run frontier-scale open models starting with Moonshot AI’s Kimi K2.5, a multimodal agentic model with a 256k-token context window and multi-turn tool-calling capability. (blog.cloudflare.com)) Cloudflare’s blog says internal usage of Kimi in its OpenCode and automated code-review pipelines drove the decision to serve the model on Workers AI, and third-party reporting puts Cloudflare’s claimed cost reduction for a single workload at about 77% (roughly $2.4M projected annual savings). (blog.cloudflare.com)) Meta’s Llama roadmap continued to raise the bar for open families when Llama 4 (Scout, Maverick, Behemoth) was released on April 5, 2025, signaling that major vendors are shipping increasingly capable open alternatives alongside closed APIs. (techcrunch.com))
