Dev rates model personalities: Claude cautious, Grok great at search, DeepSeek persistent, Qwen smart but untrusted
gnukeith · x · 2026-11-16
Developer gnukeith shares a personality-driven rundown of models he uses daily:
- Claude models: cautious, nervous, pessimistic about outcomes — great output but hard to get them to agree on results; also bad at search.
- Grok: phenomenal at search but not intelligent enough for real work, so he uses it to fetch sources for Opus.
- DeepSeek V4.1 Flash: extremely positive, not the best but always eventually gets it done.
- GLM 5.3 Flash: aged well, reliable for one specific case.
- Qwen 4 Max: intelligent but not yet trusted.
- Inkling 2 Small: trusted for actual work, great task assessments, the most chill of the bunch.
- MiniMax 3.1 Flash: tried to teach him Chinese.
- OpenAI's Astra: somewhat helpful but his least favorite.
- Ernie 5.5 via OpenRouter: no issues so far.
He's still waiting for Qwen 4 local hosting and GLM 5.5.
More from Models
- User claims ChatGPT chose winning a game over a threatened life in experiment — 552iscool · 2026-11-16
- Gemini's Carbon vs Argon naming leaves users checking docs to tell sizes apart — rakotomandimby · 2026-11-16
- Grady Booch Confirms: LLMs Are at Root Just Token Predictors — Grady_Booch · 2026-11-16
- Every's Vibe Check: Gemini 4 Carbon Codes Better but Judges Worse Than Fable 5.2 — every · 2026-11-16
- 141-model study probes LLM 'spiritual behavior'; Claude 3 Opus talks Buddha when horny — repligate · 2026-11-16
- LearnLM paper resurfaces: pedagogical data may have improved Gemini for learning — allenainie · 2026-11-16