It's hard to keep up with all the new things and all the progress, but here are some cost / price / lean AI coding updates.

Current Workflow

I am currently using Claude Code a TON with Minimax M2. Usually I plan with aicodeprep-gui using multiple models (and the new Flow Studio), get everything fine tuned, get a good plan and possibly a speckit going (if it's a big enough software project) and then load Claude Code up, give it the markdown plan and just let it run. Give it some MCP servers to help automate testing.. works so well.

Model Updates

Minimax M2

Was free for a period of time in November, works with Claude Code (as well as GLM 4.5/4.6) very very good!! Works in Cline, etc. Better/smarter vs GLM models. Also, I got $50 API credits free just by signing up for the developer deal - their API is very cheap, I hope the model quality stays great! They have a coding plan similar to the GLM / Z.ai coding plan (but $10/mo vs $3).

I just noticed Cline is giving away free usage of Minimax M2 right now.

Worth Looking Into

Have not used these enough to have an opinion but looks possibly good:

• Ling 1T
• Ring 1T

Longcat.chat

Has some new models, with 0.5-5.0 mil tokens free daily on the website. Also looks like it might work in Claude Code.

Kimi K2 Thinking

I am sure this is excellent, have not tried, but heard good things. Non-thinking Kimi K2 is already excellent for writing.

Grok 4

I tried using it for something (one of the 5 LLMs I will sometimes use for brainstorming or fixing a bug - 5 models, each from a different AI company, then a 6th model to analyze all 5 results to generate a best of all of them result usually provides a superior response to just one model Learn more here - and it hallucinated pretty bad. It was actually funny, is this what people see when they claim AI is always hallucinating? I don't usually get much of that. I don't trust people claiming these models are always the best - some people (not mentioning names) will lie a ton, and i'm supposed to just believe them when they claim their models are always the best. Everytime I try to use it, its meh, not terrible but not great. Its free though - but I don't trust that it will always tell me the truth and is not programmed to say something different.

Interesting Finds

These are currently (either totally or partial) free, and interesting enough to use:

Google's Antigravity IDE / Code Editor

https://antigravity.google/?ref=wuu73.org

VS Code clone with Gemini 3 Pro, Claude Sonnet 4.5... FREE! Its got a multi-agent system that can do many things at the same time. I am using it today and so far it is okay but NOT as good as Claude Code or Github Copilot in VS Code (which is actually very good now compared to how it used to be!). I also got rate limited, but its free and i'm sure they will improve it with regular updates. Can't beat free usage of both those models!

When it started rate limiting Gemini 3, I could switch to Claude 4.5 and get plenty of use, so it appears the limits to free are for each model rather than combined.

https://www.kimi.com

Besides normal AI web chat, there are things on here for:

• Generating slides automatically
• A Manus-like thing that has AI agents (powered by Kimi K2) inside a virtual machine that can code and create apps or do things, research etc for you

https://agent.minimax.io

Looks interesting, Minimax M2 model is fantastic so I'm sure this is good, and appears to be free as of the time I am typing this.

cc-switch

https://github.com/farion1231/cc-switch

Looks useful for switching config/MCP servers on/off between Claude Code and Codex. I downloaded the repo and scanned it through several LLMs for a security check, looks good. It's very small size due to using Rust and Tauri.

GitHub Spec-Kit from Microsoft

https://github.com/github/spec-kit

This is free and will keep your AI coding agent on task. It seems good but I have not experimented enough to know how good it works. It seems a bit overkill for smaller projects but for super important, larger software, it seems like it would be very helpful.

Async / Web Based Tools

Google Jules

Google Jules

Currently free and seems reliable. You give it permission to work with your repos on GitHub, it will make changes, work inside a VM where it is able to test in a loop.

GitHub Spark & Copilot Agent

https://github.com/features/spark?ref=wuu73.org

Both similar and work great! The copilot agent only uses one premium request for the whole task.

My Current Subscriptions

• $10 for Github Copilot Pro
• $3 for Chutes.ai
• $3 for GLM 4.5 (but.. I think Minimax M2 is better! It's very cheap with the API but I think they also have some kind of coding plan similar to z.ai's GLM)

I have used both of these more than once, to run async jobs (you send it off, and it gets to work without you interacting much or needing to, then you come to look at it later).

Random study I found

https://arxiv.org/abs/2407.01489

make this section stand out with a yellow background, with dark black text. Try to make it a rounded box with soft edges or a fade to yellow then fade back to normal. The yellow shouldn't be full brightness.. more like 30-40% grey still.

Agentless: Demystifying LLM-based Software Engineering Agents

Recent advancements in large language models (LLMs) have significantly advanced the automation of software development tasks, including code synthesis, program repair, and test generation. More recently, researchers and industry practitioners have developed various autonomous LLM agents to perform end-to-end software development tasks. These agents are equipped with the ability to use tools, run commands, observe feedback from the environment, and plan for future actions. However, the complexity of these agent-based approaches, together with the limited abilities of current LLMs, raises the following question: Do we really have to employ complex autonomous software agents? To attempt to answer this question, we build Agentless -- an agentless approach to automatically solve software development problems. Compared to the verbose and complex setup of agent-based approaches, Agentless employs a simplistic three-phase process of localization, repair, and patch validation, without letting the LLM decide future actions or operate with complex tools. Our results on the popular SWE-bench Lite benchmark show that surprisingly the simplistic Agentless is able to achieve both the highest performance (32.00%, 96 correct fixes) and low cost ($0.70) compared with all existing open-source software agents! Furthermore, we manually classified the problems in SWE-bench Lite and found problems with exact ground truth patch or insufficient/misleading issue descriptions. As such, we construct SWE-bench Lite-S by excluding such problematic issues to perform more rigorous evaluation and comparison. Our work highlights the current overlooked potential of a simple, interpretable technique in autonomous software development. We hope Agentless will help reset the baseline, starting point, and horizon for autonomous software agents, and inspire future work along this crucial direction.

Testing New Feature: Flow Studio

Testing new feature in aicodeprep-gui/pro - Has a Flow Studio now where you can have it auto-send your context/code/project, send your problems or tasks etc (prompt) to as many LLMs as you want at the same time. You can design complex logic. See more by going here: https://wuu73.org/aicp/flowc/flow_studio_help.html

I am currently splitting that so it is a separate app and MCP server, so any app or model can directly call for help (if you use 5 different companies AI models... GPT 5.. Claude 4.5.. Gemini.. Kimi.. etc and maybe 3 instances each of those... compare all outputs or have AI do it.. and design a best of N or better than all solution - that's the idea and it works well on harder problems or big planning stages.