zephyr4123/speakmore is an open-source voice input tool for Mac and Windows. According to the repository README, you press a shortcut key, speak, and press it again, and the cleaned-up sentence is typed at the cursor position. The sources for the currently confirmed information are the repository description and the README; FACTS does not record the latest release.
The tool is built so you can choose between dictation style and which model to use. The README explains that it supports 29 models plus OpenAI-compatible services, citing Gemini, OpenAI, 通义千问, DeepSeek, Kimi, 智谱 GLM, and MiniMax as examples. On-device models can run on Mac and Windows via Metal·Vulkan·CPU, so there is also a path that does not send audio externally. It offers three recognition methods side by side, and the recommended combination can be enabled all at once. After recognition, it cleans up filler expressions, repetitions, and correction utterances, supplements punctuation, and splits long content into segments and items.
However, the README contains no performance figures or benchmark tables, so there is no basis for comparing recognition quality with other tools. For cost, it only states that the recommended combination at 100 uses per day runs about 0–35 yuan per month, and does not guarantee the cost for every model or every usage condition. So it is confirmed to be a tool that offers multiple options, but quality and actual cost may vary by the model you use.