Does 5× more work for the same tokens. Your subscription lasts longer. Free-tier models actually work.
Every other agent is incentivized to burn tokens. 3code is built to save them — aggressive caching, context compaction, chunked mode that discards irrelevant data instead of carrying it forever. The result: your API balance lasts 5× longer, and you can use free-tier models for real work.
3code also supports moonshot's Kimi (both subscriptions and api), Deepseek, OpenRouter, and many others.
$ curl -fsSL https://3code.capocasa.dev/install | sh$ curl -fsSL https://3code.capocasa.dev/install | sh> irm https://3code.capocasa.dev/install.ps1 | iex$ mkdir myprojectdir
$ cd myprojectdir
$ 3code
╶──╮
╶──┤ 3code v0.6.0 the economical coding agent
╶──╯
no provider configured. let's add one. (ctrl+d to quit)
supported: deepinfra, ovh, nvidia, nebius, fireworks
api key : ************************
provider name or url : nvidia
verifying... ok
saved to ~/.config/3code/config
❯ Build me a Hello World program in NimClaude Code, Cursor, Copilot — they all optimize for capability. Nobody optimizes for cost. 3code is the only coding agent that treats your token budget as a first-class constraint.
That means your $20 subscription lasts a month instead of a week. It means you can use DeepSeek V4 or GLM-5.2 free tier and still get real work done. It means you don't have to think twice before asking a question.
We eat our own dog food — 3code is now written entirely with 3code, using GLM 5.2 on Z.ai's free tier.
There are a many, many design choices where small adjustments lead to huge savings. Then there is one unique innovation- chunked mode- that constantly extracts relevant and discards irrelevant data and keeps agentic coding with very little context- and it improves results because of intelligent discarding.
3code is written as a system program and is absolutely tiny. It always loads and responds instantly- as an AI assisted programmer you are an extremely productive individual- your coding agent should not keep you waiting, even for a few seconds.
The idea that you have to trade efficiency for ease of use is so widespread, most designs that are efficient technically will have poor usability. But it doesn't have to be that way- Enjoy efficiency that's a joy to use.
opencode sent 4.9× the non-cached input of 3code on comparable work, driven by higher tool-call volume: more file reads, more bash, more context re-sent per turn.
On most providers cached input accounts for 90%+ of token spend, so this is where efficiency pays off most.
Zero eval errors, zero unresolved patches; every timeout produced an empty patch. Full per-task breakdown, observations and methodology →