Open weight LLMs exhibit inconsistent performance across providers

A benchmark of the open-weight LLM gpt-oss-120b shows significant performance inconsistencies across different hosted providers. These variations are caused by differences in the underlying infrastructure, such as the serving framework version or the use of model compression. This creates a challenge for users who cannot rely on a model's name alone to guarantee performance. The situation highlights the need for standardized conformance suites to help verify provider implementations, and some model creators are now releasing compatibility tests to address this.