NVIDIA QUIETLY OPENED UP FREE API ACCESS TO 100+ AI MODELS, NOT JUST 70. GLM-5.1/5.2, DeepSeek V3/R1/V4 Pro, Qwen, Kimi, Nemotron, Llama, Mistral, Gemma, all through one OpenAI-compatible endpoint. Context and output sizes match the original models exactly, no artificial trimming, up to 1M tokens on models that support it. Setup: go to http://build.nvidia.com, sign up free, generate a key starting with nvapi-, valid for 6 months. Point any OpenAI-compatible client, Cursor, Cline, Aider, OpenCode, Claude Code, at base_url https://integrate.api.nvidia.com/v1. Rate limit sits around 40 requests per minute, best-effort, not guaranteed. Worth knowing before sending anything sensitive. NVIDIA's own docs don't clearly commit to zero logging, so treat this as a prototyping and learning environment, not where confidential code goes.(CyrilXBT)