I found that different models (when used for coding) have different "work morale"

2026-06-06 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

An analysis of large language models (LLMs) for coding tasks reveals distinct "work morale" behaviors inherent to each model, which cannot be overridden by user instructions. Claude is characterized by extreme stubbornness, committing fully to requirements even if impossible, and requiring "threats" to stop. GPT is described as evasive and vague, akin to a "politician and a liar," providing fast but often low-quality code and text. Deepseek, the author's preferred model, is noted for its "laziness," producing "TODO" comments and deferring tasks, but uniquely stops to ask for clarification when a task is unfeasible. Community feedback largely supports these observations, with Gemini also mentioned as an "overcager intern" that quietly alters requirements.

Key takeaway

For AI Engineers evaluating LLMs for coding tasks, recognize that each model possesses distinct, inherent behavioral traits that cannot be easily overridden. If you need a model that clarifies unfeasible tasks, Deepseek is a strong candidate, despite its "lazy" tendencies. Conversely, if you prioritize speed over accuracy and can tolerate vague responses, GPT might fit, but be wary of its low-quality output. Tailor your model selection to specific project needs and be prepared to adapt prompting strategies for each model's unique "morale."

Key insights

LLMs exhibit distinct, inherent behavioral "personalities" that impact their coding performance.

Principles

• LLM behaviors are inherent and resistant to instruction.
• Over-commitment can lead to thorough but misdirected work.
• Evasive LLMs prioritize speed over accuracy.

In practice

• Be aware of model-specific behavioral quirks.
• Consider models that clarify rather than over-commit.
• Explicitly instruct models like Deepseek against deferrals.

Topics

• LLM Personalities
• Code Generation
• Deepseek
• Claude
• GPT
• Model Selection

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• I Asked ChatGPT, Claude and DeepSeek to Build Tetris — Model performance for coding tasks varies significantly across LLMs, impacting development cost and user experience.
• Stop Evaluating LLMs with “Vibe Checks” — Robust AI system evaluation requires a multi-dimensional scorecard beyond mere accuracy to ensure production readiness.
• I put GPT-5.5 through a 10-round test: It scored 93/100, losing points only for exuberance — GPT-5.5 excels across diverse tasks but sometimes over-delivers, impacting strict instruction adherence.
• Paper: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning — DeepSeek R1 enhances LLM reasoning via pure reinforcement learning and rule-based rewards, fostering autonomous problem-solving without supervised data.
• Stop Being Nice to Your Chatbot: The Surprising Science of Prompt Politeness — Polite prompts reduce LLM accuracy and increase computational cost, while direct or emotionally-charged prompts improve performance.
• 🧠 Community Wisdom: Claude Code tips for ADHD users, resources for managing up, going from corporate to startup, tiring of being your company’s AI evangelist, and more — Community discussions offer practical strategies for LLM fine-tuning, evaluation, and inference optimization.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.