How To Optimize Your Usage: The Best AI Models to Use, version 3.0

[Artemonim]

I tried Gemini 3 Flash again, for a less important part of the project. I’m a little afraid of its randomness, but for quickly working on unimportant things it’s really good. When it stuttered, I asked GPT-5.2 to double-check/debug it.

[Artemonim]

@Artemonim

• GPT-5.2 XHigh: main developer
• GPT-5.1 Codex Max XHigh: for fast and simple tasks
• GPT-5.2 Codex: test writer
• Gemini 3 Pro/Flash: for discussions and chatter in Ai Studio without project context

I tried Cursor’s GPT-5.2 XHigh and GPT-5.1 Codex Max XHigh for writing tests – and I got the feeling I’d spend more money on tests than on the code itself.

My current list:
(GPT-5.2 XHigh (~= 5.2 Codex XHigh?) ~=$$ Claude Opus 4.5 Thinking) $> Gemini 3 Flash $> GPT-5 Mini > Grok Code Fast

---

(Continuously Updated) My Real-Time Review of Cursor Subagents

Testing subagents from the Cursor Nightly build. I tried building a search agent based on Grok Code Fast. The primary agent was GPT-5.2 XHigh. Conclusion: GPT-5.2 XHigh fills context more efficiently when performing the search itself than when reading context based on instructions from others. However, it can be suggested to use a search agent based on GCF, Gemini 3 Flash, or Claude Haiku for the second search pass. I’ll let you know if I have any further details on this subagent.

[Artemonim]

Agent Enforcer 2: Enforce the Quality of your Project

Deprecated: Agent Enforcer Open Beta: Enforce the Quality of Agent Code A reference architecture for building robust, language-agnostic local CI systems. Agent Enforcer 2 is not a tool you install — it’s a concept and blueprint for implementing local CI in your projects. It documents battle-tested patterns for orchestrating code quality checks, managing caches, detecting hangs, and producing structured reports. Last summer I tried to publish my approach to building a more automated dev wor…

GitHub - Artemonim/AgentEnforcer2: A reference architecture for building robust,...

[Wearetheprojekt]

I am new to cursor, and I want to know if it is advisable to use Auto mode for selecting agents?
Because I have been reading around and seems like everyone prefer to select there own model. I do know that auto mode selects the best model for the task, so I am wondering why there is no advise telling people to just keep the mode in auto if you want to avoid burning tokens and still want to get good work done. @Artemonim

[Artemonim]

With Auto you have less control. If you try different models and don’t notice a difference, then you’ll probably just need to stick with the cheaper option. But if you do notice a difference, you’ll choose models based on specific needs.

[Zec (BaoYuanZi)]

GPT 5.3 and Opus 4.6 have been released. I’m curious whether you would still use GPT 5.3. I haven’t tried it yet and I’m interested in its capabilities — I’ve only used Codex before.