• Added Fluid-1 Mini Research Preview, a second smaller AI enhancement model. Half the size of Fluid-1 at 1.5 GB, downloads and runs faster and is currently more capable than Fluid-1 for dictation tasks.
• Added Learn from your edits, on by default. Correct a word after a dictation and FluidVoice adds it to your custom dictionary, with a card to undo. Only distinctive words, never everyday ones. Nothing is sent anywhere and password fields are never read.
• Rebuilt the Stats page around time saved, personal bests, a seven-day activity strip, and the apps you dictate into most.
• Made type sizes and spacing consistent across every screen.
• Fixed long dictations being cut short after roughly 1,200 characters.
• Fixed older GPUs attempting an unsupported CUDA runtime instead of falling back to the processor.
• Fixed the Stats seven-day chart leaving out the current day, and empty days drawing as full blocks.