Yi-Large has credible historical capability evidence, but the public record is dated and incomplete. The July 2024 Arena-Hard-Auto snapshot puts Yi-Large at 63.70, with Yi-Large-preview at 71.48; this supports a mid-tier general reasoning score rather than present-frontier positioning. Its reported HumanEval basic 0.524 and EvalPlus 0.652 indicate usable baseline coding ability, but do not establish repository-agent performance. It is materially below Claude Opus 4.8 and GPT-5.5 on evidenced frontier reasoning and coding, while its general profile is closer to Hermes 4.3 36B than GLM-4.6. Official sources substantiate hosted chat/completions access and OpenAI-compatible integration, supporting a workable developer-experience score. They do not substantiate image, audio, tool-use, structured-output, context-window, rate-limit, regional, latency, or reliability claims; multimodal/I/O and availability therefore score below text-API comparables such as GPT-5.6 Terra. The Yi GitHub repository is a positive public developer signal, but it is not a model-specific API reference for Yi-Large. Cost and pricing are deliberately conservative: the official material says flexible pricing but supplies no rates, billing units, or quotas. DeepSWE lists no Yi-Large result; no SWE-bench evidence was found; LiveCodeBench has no Yi row; and the supplied Artificial Analysis source provides no Yi-Large intelligence, price, or speed data. Arena-Hard is useful public preference evidence but historical, so it should not be read as a current competitive ranking.
