MLPerf Inference v6.0 introduces GPT-OSS 120B, a new open-weight LLM benchmark, plus a DeepSeek-R1 interactive scenario with support for speculative decoding.