Muse-glimmer-30b punches above its weight in creative writing, outclassing larger models
spanielrassler · reddit · 2026-09-11
After noticing Muse-glimmer-30b ranks alongside much larger frontier models on EQ-Bench Creative Writing and Hemingway-bench, the author tested it for style emulation and was impressed.
- Method: Asked it to imitate Henry Miller, David Sedaris and Stephen King — informal testing, but glimmer beat the comparison model in every case.
- Head-to-head: On "write a humorous paragraph in the style of David Sedaris," glimmer nailed the self-deprecating, detail-stacking voice on the first try (author actually laughed), while qwen3.8-27b's attempt fell flat.
- Observations: No system prompt was used; the author suspects this agentic-coding-trained model could steer even better with one. Community finetunes are mostly abliterated variants — surprisingly few writing-focused finetunes despite this performance.
A concrete, sample-backed reference for anyone watching local small models' creative-writing ability.
More from Models
- CursorBench 4.0 rolls out with harder, longer-horizon coding tasks, scores drop — StringChaos · 2026-09-11
- Developer: Astra is a good model — jxnlco · 2026-09-11
- New Model Hits Opus-Level Benchmarks at Wild Efficiency, RL Infra Details Emerge — nrehiew_ · 2026-09-11
- DeepSeek V4.1 Flash notes: how obsessing over KV cache compression yields a hyper-efficient frontier model — nrehiew_ · 2026-09-11
- Model with reasoning off finds bugs Opus 5 missed — cephaloform · 2026-09-11
- DeepSeek V4.1 Flash goes live on Baseten with 1M context and vision support — baseten · 2026-09-11