Code Generation
6 articles tagged with this topic
Two Years On, AI Coding Still Has 4 Pitfalls — Bottleneck Was Never the Model
A Juejin roundup lists AI coding's 4 most common pitfalls. Two years on, code comes out faster — but judging whether it's correct still has no shortcu
AI Code Generation 10x Speedup Is Real — But Only After You Lock Down the Rules
Framework author hit 10x efficiency with LLM code gen — but admits templates, not AI, do the real work. A must-read for any AI efficiency team.
Xiaomi MiMo Wastes 6x Compute on Junk Code; LLMs Shift to Delivery Efficiency
Xiaomi MiMo burned 6x compute for junk code while DeepSeek excelled. Benchmarks no longer reflect true dev capability; focus on delivery and costs.
Devstral Small 2 Breaks 80% Code Benchmark — Mistral May Be Seriously Underrated
Developer's custom benchmark: Mistral's Devstral Small 2 scores 80%+ on 8 code tasks—first local model to beat multiple closed-source rivals.
GLM 5.1 Dominates Open-Source Code Arena: China's Programming AI Inflection Point
Zhipu's GLM 5.1 topping open-source code rankings signals that low-cost programming AI is within reach, with software outsourcing and IT service prici
Qwen3.5 vs Gemma4 vs Cloud LLMs: Python Turtle Drawing Benchmark
A Reddit user benchmarks local and cloud LLMs on Python turtle graphics, revealing Gemma4 and Gemini share visual style.