Summary
Tests on four B200 GPUs found matching image-tokenisation fingerprints and token counts, while a reconstructed hidden system prompt produced a 75-token delta and explicitly named ox-alpha. The identification is an analysis rather than a confirmed disclosure from Baseten, so the model’s provenance remains uncertain.
Why it matters
The case illustrates how model identity and provenance can be inferred from serving behaviour even when providers release systems anonymously.
Related coverage
- Ox Alpha reports an 80% pass rate on the full 113-task DeepSWE run
- GLM-5.3 doubles its reported KernelBench-Mega speedup
- Microsoft receives first production Vera Rubin systems as Nvidia ramps the platform