Official Nous materials establish a 36B open-weight hybrid-reasoning model with 128K context, structured outputs, GGUF variants, and local-deployment support. Microsoft Foundry additionally documents a vLLM-backed, OpenAI-compatible Chat Completions deployment. These are practical strengths, but supplied evidence contains no independently reported quality, coding, latency, or preference result. Technical and reasoning scores therefore sit around the GLM-4.6 anchor rather than near DeepSeek-V3.2 or frontier hosted models. Coding is scored materially below DeepSeek-V3.2 (84) and Claude Opus 4.8 (95): DeepSWE does not list Hermes 4.3 36B, LiveCodeBench has no matching entry, and no SWE-bench result was found. Developer experience exceeds GLM-4.6's 73 on the documented GGUF and OpenAI-compatible deployment paths, but remains below Gemma 3's 86 because the supplied record provides less ecosystem and workflow evidence. Multimodal I/O is below Gemma 3 because official materials here substantiate text capabilities only. Open weights can make cost attractive for operators able to self-host, but 36B inference hardware requirements and absent hosted rates limit that advantage. Pricing clarity and availability are weak: Nous Portal marks the model unsupported, unavailable through its API, and priced N/A. Public adoption is modestly scored from its established distribution channels, not benchmark or usage statistics. Evidence quality is constrained by missing independent evaluations and an unspecified license.
