GLM improved its inference. Engineers still set the goal
Z.ai reports 3× serving throughput with help from a GLM coding agent. I examine what improved, what humans controlled, and what developers can reuse.
coding-agents
Every article tagged coding-agents, newest first.
Codex CLI adds voice. The recovery fixes matter more
Codex CLI 0.155 adds experimental voice and safer worktree cleanup. I explain the fixes for lost prompts, restarted sessions and the next-day hotfix.
DeepSeek V4.1 Flash: low API prices, huge open weights
DeepSeek V4.1 Flash activates 8B parameters on input, but its full weights are enormous. I explain the API prices, coding scores and server requirements.
Fable 5.1 beats Fable 5. Opus is harder to replace
Fable 5.1 leads its predecessor on long agent tasks, but Opus 5 stays close at half the base token price. I checked the benchmarks and first reports.
I'm torn between Claude and Codex
Codex gives me more usage, cleaner frontend work, and a smoother app. Claude still finishes my large multi-agent features in hours instead of all day.
Opus 5 is not as bad as the internet says
After five weeks of daily use, I think Opus 5 is much better than its online reputation. Clear plans and smaller tasks reveal a strong coding model.
Cursor didn't remove the IDE. Here's how to get it back
Cursor's Agents Window can open first, but Editor Window remains supported. I map the commands and startup setting that put code back in front of you.
Why similar AI benchmark scores can hide different models
DeepSWE puts Luna Max 2.2 points behind Sol High at roughly one-sixth the attempt cost. I explain why the models can still feel far apart in repository work.
My coding agents forget the repo. I built them a map
I built a four-layer context management system that routes coding agents to current facts. It works, but stale documentation is still the hard part.
GPT-5.6 vs Claude 5: why the benchmarks disagree
I compared Sol, Terra, Opus 5 and Fable 5 across coding benchmarks. The winner changes with the task, effort setting, agent setup and budget.
Kimi K3 matches GPT-5.6. It takes 4.6 times as long
Kimi K3 ties GPT-5.6 medium but takes 4.6 times as long. GLM-5.2 is cheap per token yet costly per task. I checked where both models still win.