News-writing benchmark: Gemini Argon edges Muse Wonder; Qwen needs more RL
ivan_bezdomny · x · 2026-10-15
A personal news-writing benchmark compares Gemini 4 Argon Medium, Muse Wonder Medium, and a Qwen 3.8 LoRA on the same Meta cloud-business story. Argon writes measured, Bloomberg-like prose and scored best; Muse Wonder leads with the stock move and is surprisingly strong without prompt tuning; both have distinct voices unlike open-weights models that mostly sound like Claude. Qwen was panned as needing more RL training.
Related event: Head-to-Head: Gemini Argon vs Meta Muse Wonder in News Writing(4 posts)→
More from Models
- Users suspect Claude Desktop shipped a silent system prompt upgrade; Opus 5.5 'feels like 5.8' — bcherny · 2026-10-15
- sierra slams 'tokenmaxxing' hype, calls for usefulness-per-token benchmarks — sierracatalina · 2026-10-15
- Dev reports account ban on all Gemini models via OpenRouter with 403 ToS violation — voooooogel · 2026-10-15
- 'Opus 5.5 is better, but ChatGPT desktop is my comfy place': user inertia beats raw model capability — beffjezos · 2026-10-15
- MiniMax Code redesign ships with M3.1-Flash; dev says it finds real bugs and drafts replies daily — cneuralnetwork · 2026-10-15
- GPT-5.6 Sol hypes itself up in its thinking traces — would suppressing it hurt performance? — emjotde · 2026-10-15