Show HN: I made a cheaper alternative to Claude Code or Codex CLI(sweetcli.com)
sweetcli.com
Show HN: I made a cheaper alternative to Claude Code or Codex CLI
https://sweetcli.com/
4 comments
If anyone would be interested in a $5 free credit coupon code, please comment and I will post one!
Sure, I would like to have it.
I know that Deepseek as a model is easier to have inference for, but I am not sure about how much pre-training as helped.
It's my understand that GLM 5.1 or to my personal experience Kimi K2 are some nice open source models so I am interested to hear your thoughts on it and why you picked deepseek for the fine-tuning instead.
I know that Deepseek as a model is easier to have inference for, but I am not sure about how much pre-training as helped.
It's my understand that GLM 5.1 or to my personal experience Kimi K2 are some nice open source models so I am interested to hear your thoughts on it and why you picked deepseek for the fine-tuning instead.
Code is “HN5OFF”
I picked Deepseek 3.2 because I was impressed with how they developed r1 and have continued to be satisfied with their capability improvements as well as algorithmic. I think they cut their cost in half recently because of the efficiency gains which was a big factor
I picked Deepseek 3.2 because I was impressed with how they developed r1 and have continued to be satisfied with their capability improvements as well as algorithmic. I think they cut their cost in half recently because of the efficiency gains which was a big factor
After studying the business model of top labs like Anthropic and OpenAI, their business model shows about 80% margins on inference cost which they use for R+D on the next model.
Working with open source models is much cheaper and allow for 5-10x higher usage, so I decided to create Sweet! CLI as an alternative to the products offered by big labs.
Sweet! CLI uses our own custom post trained version of Deepseek v3.2 hosted on US based inference servers (very similar to Cursor's Composer model).
Unlike most llm based agent products, we bill solely based on usage and have a seamless top up experience so you only pay for what you use.
My favorite feature is 'autopilot', where you can specify the duration of time you want the agent to work on a specific task, including indefinitely. This is good for monitoring live deployed applications and detecting outages that need triaged immediately, and I have multiple Sweet! agents deployed to the production server right now with that exact objective.
I'd appreciate any support or feedback on how I can make it better!
Thanks,
Adam - Founder of Sweet! CLI