Reddit
A 0.8B Apache-2.0 fine-tune ties GPT-5.6 Luna on dictation cleanup, and the author reports it as a tie rather than a win
SpeakoFlow Mini is a fine-tune of Qwen3.5-0.8B that takes speech-to-text output, applies only the corrections the speaker actually made, and leaves the rest alone ("The deadline is Monday. Scratch that. The deadline is Wednesday." → "The deadline is Wednesday."). On the author's English-only benchmark it scored 70.7% against GPT-5.6 Luna's 65.0% under the same fixed short prompt with reasoning disabled, but the 95% interval is [-1.5, +12.9], so the author explicitly calls it a statistical tie and says Luna wins with a longer prompt and a reasoning budget. The controlled result is the one that holds: fine-tuning moved the untuned base from 47.3% to 70.7%, +23.4 points at [+16.3, +30.3].
Source
↳ Follow the thread