NVIDIA ran someone else's open model on its newest rack, positioning that hardware as the floor for agentic coding.
What happened
An NVIDIA developer post walks through experimenting with the 176B Qwen3.8-Flash-Next model on GB300 NVL72 for agentic coding. The cluster is thin — one official post and one academic paper — with no cross-vendor pickup yet.
Why it matters
Showcasing a third-party open model on your latest rack reads as the GPU side claiming the substrate anyone's model runs on. Still, this is a single experiment write-up, not evidence of deployment or GA; that remains unconfirmed.
What to watch
Third-party benchmarks on the same configuration, and reporting on actual GB300 installs. Watch whether agentic-coding evaluation shifts from throughput to cost.