Multiple $20 AI Plans Are Better Than a Single $100 AI Plan

Multiple $20 AI Plans Are Better Than a Single $100 AI Plan - Abishek Muthian

Abishek Muthian

Multiple $20 AI Plans Are Better Than a Single $100 AI Plan

Posted at — Jul 1, 2026 by Abishek Muthian

After using the $200 and $100 Anthropic monthly plans for Claude, I have now settled on $20 Anthropic + $20 OpenAI + $10 OpenCode Go monthly plans for coding and design work. I also use local AI occasionally for non-coding-related tasks.

By using multiple entry-level AI plans from different providers, I can assign different tasks to different models based on their strengths and also hedge against service disruption by not relying on a single provider.

This is how I do it:

IDE

I use the Zed[1] IDE for coding. Its Agent Panel offers seamless integration for top AI coding agents. It has a proper terminal and LSP support for major programming languages.

Despite using its own Rust-based nascent GPUI[2] framework, it has been more stable than VS Code has ever been for me and more performant than any non-TUI-based IDE.

I use Zed across Linux, macOS, and Windows, and my experience with it on all three platforms has been really good.

Major caveats include the lack of accessibility support and the lack of blessings from SDK developers like Flutter; but since Flutter has a robust CLI, I've had no issues building Flutter apps in Zed.

But if your SDK/framework relies extensively on a VS Code extension, then Zed may not work for you right now.

Agents

Zed has support for major AI coding agents except Gemini, since Google went all walled-garden with Antigravity[3].

These are the agents I use:

Claude Code[4] with the latest Opus model at default effort for planning.

Codex[5] with the latest ChatGPT at medium/high effort for implementation of the plan.

OpenCode Go[6] with GLM 5.x, DeepSeek V4 Flash, and Qwen 3.6 at high effort for maintenance and improvements to my existing codebases.

Note: OpenCode seems to be run entirely by AI, as there are numerous billing-related issues and zero support. I lost money from a double payment for their subscription, so I re-subscribe every month.

MCP

I have these MCP servers across all my Zed setups:

Fetch[7] for fetching web content.

Brave Search[8] for searching the web.

Puppeteer[9] for web scraping.

Excalidraw[10] for making architectural diagrams pretty.

Local AI

I have an RTX 4090 laptop and an M4 Mac Mini. After experimenting with local AI for coding for the past 3 years[11], I've come to the conclusion that LLMs on local hardware for coding are not worth the cost.

The latest open-weight coding models are competitive with proprietary models[12]. To run them locally at acceptable quality, I would need to invest in buying better hardware than my already small-car-priced laptop.

I can't justify such levels of consumption when I can get the same work done for $50/month.

I do keep a few quantized LLMs that my machine can run locally via llama.cpp[13], for curiosity and backup; after all, these models are a knowledge corpus, albeit probabilistic in nature.

Non-coding SLMs (Small Language Models)[14], on the other hand, are a totally different story. It's unfortunate that coding LLMs on local hardware get all the attention when purpose-built SLMs can solve many problems while being capable of running even on mid-tier CPUs.

I use Krita with the AI diffusion plugin[15] for occasional photo edits.

I use Unsloth Studio[16] for fine-tuning small language models.

I use Pinokio[17] for experimenting with TTS, background removal, image upscaling, small animations[18], etc.

I will write more about the SLMs I use in a future post, as this post is specifically about coding models.

Conclusion

This is the AI coding setup that works for me now. I'm in no way claiming that this is the setup everyone should follow.

[1] https://zed.dev/

[2] https://gpui.rs/

[3] https://github.com/google-antigravity/antigravity-cli/issues/31

[4] https://claude.com/product/claude-code

[5] https://chatgpt.com/codex/

[6] https://opencode.ai/go

[7] https://github.com/modelcontextprotocol/servers/tree/main/src/fetch

[8] Available as Zed official extension.

[9] Available as Zed official extension.

[10] github.com/excalidraw/excalidraw-mcp

[11] https://abishekmuthian.com/how-i-run-llms-locally/

[12] https://semgrep.dev/blog/2026/we-have-mythos-at-home-glm-52-beats-claude-in-our-cyber-benchmarks/

[13] https://llama.app/

[14] https://abishekmuthian.com/building-an-offline-ai-stoic-chatbot/

[15] https://kritaaidiffusion.com/

[16] https://unsloth.ai/docs/new/studio

[17] https://pinokio.co/

[18] https://abishekmuthian.com/ai-for-generating-2d-animations/