Use Modular Models in Big-AGI.

Open models, served fast, on prepaid credit.

Modular Cloud is the hosted endpoint of the MAX and Mojo team, now part of Qualcomm: open-weights models on NVIDIA and AMD GPUs, one sk-mod- key, pay-per-token from a prepaid balance. In our spot checks of August 13, 2026 - two ~200-token streaming runs per host from a US connection, not benchmarks - it was the fastest host of every model we compared that day, with MiniMax M3 at ~238 tokens per second where the model's own API measured 64.

The fine print, verified the same day: models are served as NVIDIA NVFP4 4-bit quantized builds (disclosed in the response's model field; our small task checks showed no quality difference against full-precision hosts), credit must be loaded before the first answer, and the same Big-AGI vendor also connects a free self-hosted MAX server.

Running Modular in Big-AGI

Add your Modular API key and run open models on Modular Cloud at their own per-token rates. Big-AGI adds no markup and no intermediary: billing runs directly between you and Modular, from the prepaid balance on your Modular account. The model list is read live, so when Modular adds a model - they have shipped day-zero releases before - it appears in your picker on the next refresh, with no app update.

One vendor, two machines

The same Big-AGI vendor connects two very different endpoints, and an Endpoint chip picks between them. Modular Cloud is the managed side: shared endpoints on NVIDIA and AMD GPUs, one key, pay per token. MAX self-hosted points the vendor at a MAX server you run yourself - the container is free, runs on NVIDIA, AMD and Apple Silicon, and speaks the same protocol, so whatever it serves shows up in the list automatically. Both can exist side by side, the cloud for the big MoE models and your own machine for everything you want local.

Your key and your data

Modular's API does not accept direct browser calls, so requests route through the Big-AGI fast edge servers - there is no Direct Connection toggle on this vendor. Your key stays stored in your browser and travels only to make your requests. Chats are stored locally first and sync only if you turn it on. The AI Inspector shows the exact request, the token counts, and a cost estimate built from Modular's published rates - the estimate never enforces anything; the prepaid balance at the Modular console does.

Modular in Beam

The models Modular serves - the same open weights available from other hosts - streamed faster here than anywhere else we tried them in our August 2026 checks. That pace suits Beam: put a Modular-served model next to Claude, GPT, and Gemini, let them answer in parallel, then fuse, cross-check, or synthesize the results. One prepaid Modular balance covers several distinct models in the scatter, and parallel runs use more tokens than a single chat.