When I uploaded this site, a couple of weeks ago, I took a static HTML Sonnet 4.6 generated for me, installed astro, pasted that code in the project, and served it as the homepage. Then I added an “About” section, created the blog, and called it a day.
There where various bugs I was lazy to fix, like inconsistent styling between sections.
This is actually a non-paid shout-out to OpenCode for their “Go” plan, for giving a very generous service using some smaller frontier models, with (mostly) great speed.
What’s OpenCode Go
It’s actually a prepaid service, where you get a lot of usage for $10/month (with a promo of $5 for the first month). Right now the service offers:
- GLM-5
- Kimi K2.5
- MiMo-V2-Pro
- MiMo-V2-Omni
- MiniMax M2.5
- MiniMax M2.7
The limits are:
- $12 of tokens per 5h window
- $30 of tokens per week
- $60 of tokens per month.
I was using MiniMax M2.7 through Kilo Gateway, and I was very impressed with its capabilities. So when I stumbled upon M2.7 on OpenCode Go, for $10/month, I was like “Sign me up”! It is usually very quick (a few hours ago I had performance issues, but I’m not sure if it was on my end, or on their end).
You can also use the free models of OpenCode Zen:
- Big Pickle (limited-time stealth model)
- Qwen3.6 Plus Free
- Nemotron 3 Super Free
- MiniMax M2.5 Free
- GPT 5 Nano
What did M2.7 helped me achieve?
I wanted to split the homepage into 4 pages:
The intro card, the about page, the projects page and the contact page. So I made a plan: Ask M2.7 do it for me. Clear instructions like “I want to move the About Me and Learning Journey sections from the homepage to the ‘About me’ page”. Then, I did the same for everything else. I had to do some changes by hand, but no biggie.
In less than 10 minutes, I went from a 2-page site (+ blog), to a 4-page site (+ blog). I used less than 1% of the 5h window for that job. I’m really happy with the plan and the model. Now I’m waiting for the Qwen 3.6 Plus model to be released, and hopefully OpenCode will add it to their Go service as well! If not, I always have some openrouter.ai balance!
Why not use Claude, GPT, etc?
To be honest, I was always amazed by the Chinese models (how efficient they have grown to be). Compared to the Big 2 (Claude and ChatGPT), I think I’m getting a better value for that, even on API pricing. Through the Kilo Gateway I’ve used $39 since February 13th, and less than a third was from Qwen, MiniMax and other big chinese models. 171 million tokens (about 3/5ths were on Chinese models), in 5400 requests.
I’m still going to use Anthropic and OpenAI models when needed (a large codebase will actually cost less to analyze through Claude than MiniMax M2.7, because they haven’t fully reached the level of US-based models), but they are getting there!