Daun Lee

Free cloud credits, real LLM work

2026-11-04 · View on LinkedIn

Free cloud credits are hard to spend. Then I pointed a frontier LLM at $300 of them, and they were gone in about 3 days.

Cloud providers hand out free credits to test with, but I was at a loss where to spend them. With an LLM, it is a different story, because it applies to all the computer work I do. I found out that you can spend them on frontier models, although I do not know how long this will stay open.

Why I tried

I was on Anthropic Pro. Building websites and course materials burned so many tokens that I considered Max. Then a Google course request came in. If I was going to teach Google, refusing to use Gemini felt like a contradiction. But I did not want a flimsy model, and I was not ready to subscribe to Gemini yet. So I spent the free $300 on Gemini 4 Argon.

What I found, after over a year with Anthropic's top models

This is my impression, not a test. I did not give the models the same task.

  1. Gemini 4 Argon: slower than Opus 5.5, but the research was huge and thorough. It did about 80% of my course materials site. Opus felt like a practitioner. Argon felt like a scholar.
  2. GPT-6 Astra: previous OpenAI models felt about 3% short to me. Astra felt sharp and really got things solved. I asked it to set up agy, the Antigravity CLI, with my Google Cloud project and API key, and it did. OpenAI's model set up Google's model. It often opened my printer settings, and I felt a slight fear that it might overstep.

To me, only Argon, Astra, and Fable or Opus really think. Fable, which I tried in a Pro promotion, was too document-bound to work with, but I was really impressed by how tightly it stuck to its sources. Then I was blocked. Opus 5.5 finds the right line. The rest, like Gemini 3.8 Flash, are great at simple Q&A, but they just predict the next word as they go. They do not stop, look back, and think. Flash sometimes skipped verifying the latest information.

Today: Argon on Google AI Pro, Astra on Azure credit via Visual Studio, Opus 5.5 on Anthropic Pro.

The LLMs that cloud providers offer can be used for real work on my local machine.

Saving compute is fair, but I was locked in to Opus, and free credits are what got me to try Gemini. Tight limits on frontier credits and quotas can mean missing prospective customers.

How to try them for free

  1. Gemini 4 Argon: it is not on the AI Studio free tier, but it is on Gemini Enterprise Agent Platform, formerly Vertex AI. New Google Cloud users get a $300 Free Trial credit for 90 days. In agy, I signed in with my Google Cloud project, not my Google account.
  2. GPT-6 Astra: Visual Studio Enterprise subscribers get a monthly Azure credit of $150. I deployed the model in Microsoft Foundry and registered the key in Copilot CLI. A new Azure free account gets $200 for 30 days, but I have not tested it.
  3. Claude Opus 5.5: also offered on Azure and Google Cloud, but as a partner model, so the free credits do not cover it.
Free cloud credits, real LLM work
Free cloud credits, real LLM work
Free cloud credits, real LLM work
Free cloud credits, real LLM work
Free cloud credits, real LLM work
Free cloud credits, real LLM work

Download the slides (PDF)

#KoreaMCT#AX#AITransformation#AIAgents#GoogleCloud#Azure
← All posts