NVIDIA AI Updates: April 12, 2026

1. MiniMax M2.7 Launches on NVIDIA Platforms for Scalable Agentic Workflows

NVIDIA. NVIDIA’s developer blog highlighted MiniMax M2.7, a sparse mixture-of-experts model with 230 billion total parameters built for agentic workflows, complex reasoning, and software engineering. M2.7 features a 204,800-token context window, outperforms Claude Opus 4.6 on the SWE-Pro benchmark, runs at approximately 100 tokens per second, and costs up to 50x less than comparable frontier models. Open weights are available through NVIDIA NIM and the open-source inference ecosystem. Source