Grandet provides an agent skill that compares model API prices to find the cheapest route, and an anonymous MCP server. The 2026-10-27 commit 8994f24 adds 118 lines to README.md and README.zh-CN.md (59 lines in each), documenting the procedure for installing the skill by cloning with a tag, together with a build history. All builds keep the product version 0.3.0 and are distinguished by build ID, and the tag format is skill-v0.3.0-build.<build>. The README states that when you clone a tag, the content hash of the installed folder becomes identical to the release ZIP.
The newly added Versions section gathers, in a single table, the publication time, the changes, and the test results compared with the previous build for each build. Starting from 20261017, every build runs the same 48 questions with deepseek-v4.1-flash and glm-5.3 in the Hermes agent, and is marked PASS if all goals are met and HOLD if some are missed. The 20261018b(j2) that the website installs contains a price snapshot (469 models, 174 free-credit offerings) for when grandet.ai cannot be reached, and is recorded as PASS 18/18, skill usage 96/96, and citation match rate rising from 97.4% to 100%. On the other hand, 20261018(i8) handles user-specified site comparison across 20 board rows, and switches to SkillHub when there is a 30-second delay.
However, all of these figures are self-reported records left by the repository operator and are not independent verification. There are also builds that ended in HOLD (20261017b is 11/13, 20261017 is 10/13), and builds before 20261016d may fail to install with EPERM on native Windows, so the folder has to be copied manually. The price snapshot is as of 2026-10-18 and the site has changed since then, so older builds are highly likely to require pages or fields that have already disappeared. Cloning a tag lets you reproduce exactly the same build as the website, but which build to choose must be judged within the scope of these caveats.