50x Cheaper Than Claude - But Can It Actually Code?

Tech With Tim
Tech With TimApr 7, 2026

Why It Matters

MiniMax M 2.7 proves that AI can autonomously improve its own tooling, delivering top‑tier coding performance at consumer‑grade pricing, which could democratize access to advanced development assistants and spur a new wave of self‑optimizing AI systems.

Key Takeaways

  • MiniMax M2.7 built its own training infrastructure autonomously
  • Model achieved 30% benchmark improvement after 100 self‑optimization cycles
  • Scored 0.719 gold medals, rivaling GPT‑5.4 and Opus‑4.6
  • Demonstrated 97% tool‑call compliance versus average 74% in complex tasks
  • Available via token plans, offering multimodal features at $20/month

Summary

The video introduces MiniMax M 2.7, an AI coding model that claims to be up to 50 times cheaper than Claude while delivering frontier‑level performance. The presenter highlights that the model not only writes and tests code autonomously but also engineered the very system used to train its successor.

MiniMax’s novelty lies in having the model construct its own research‑agent harness—data pipelines, CI testing, and evaluation tools—without human intervention. Over more than 100 self‑optimization rounds the system improved internal benchmarks by 30 percent, earned a 0.719 gold‑medal score on the OpenAI ML‑Bench competition, and placed just behind GPT‑5.4 and Opus‑4.6. In tool‑rich environments it achieved a 97 percent compliance rate, far exceeding the typical 74 percent.

The creator demonstrates the model in real‑world demos, building a Next.js system‑monitoring dashboard in a single prompt, then iterating UI tweaks, alerts, and a full data‑layer refactor via WebSocket. The setup is accessible through an API or a token‑plan subscription—$20 per month for 4,500 requests every five hours, with multimodal capabilities included.

If MiniMax’s open‑weight release materializes, developers gain a high‑performing, self‑improving coding assistant at a fraction of current costs, potentially accelerating autonomous AI research and reshaping the economics of software development tools.

Original Description

MiniMax 2.7 is an AI model writing code, testing it, figuring out what went wrong, and fixing it with zero human involvement. The reason I'm covering this is not just because it's a good coding model, it's because this model helped build itself. I'm going to break down what this actually means, how well the model performs, and then I'm going to use it to build a bunch of real projects.
Want to make real money with coding? I share high-signal insights on careers, monetization, and leverage in my free newsletter. Join here and get my guide How to Make Money With Coding instantly: https://techwithtim.net/newsletter
🚀 Tools I Use
Get 10% off with code techwithtim
Wispr Flow (Best AI Dictation): https://ref.wisprflow.ai/techwithtim
🎞 Video Resources 🎞
⏳ Timestamps ⏳
00:00 | MiniMax M2.7
00:44 | Self Evolution and Benchmarks
03:22 | API Platform & Cost
05:20 | Setup in Claude Code
06:58 | Setup in Cursor
08:18 | One Shot Prompting
09:30 | Full App Build
14:35 | Final Thoughts
Hashtags
#MiniMax #Claude #AIAgents
UAE Media License Number: 3635141

Comments

Want to join the conversation?

Loading comments...