Coding agents · CA6
Run a coding model locally or cheaply
15 candidates. None have been tested, so this list is alphabetical, not a rank.
Standard test task
Run a coding model on a 16GB laptop or via cheap API; complete the CA1 task; record cost/speed.
Not tested yet. A dash is an empty score, not a zero. We do not estimate results, prices, or ranks.
Not tested
Aider
Terminal pair programmer with git integration
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
Cline
Autonomous coding agent in VS Code
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
Continue
Open-source IDE assistant, any model
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
Gemini CLI
Google's open-source terminal agent
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
Groq
Very fast inference for open models
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
Kilo Code
Open-source VS Code/JetBrains coding agent
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
LM Studio
Desktop app to run local models
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
Mistral (Codestral)
Codestral/Devstral coding models
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
Ollama
Run open models locally
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
opencode
Open-source terminal coding agent
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
OpenHands
Open-source autonomous coding agent
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
OpenRouter
One API for hundreds of models
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
Qwen Code
Terminal agent for Qwen coder models
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
Roo Code
VS Code agent with custom modes
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Not tested
Tabby
Self-hosted coding assistant
- Job resultWeight 40
- —
- Speed and effortWeight 15
- —
- Price and free tierWeight 15
- —
- Ease of startWeight 10
- —
- Reliability and trustWeight 10
- —
- Community signalWeight 10
- —
- Last tested
- Not tested yet
- Evidence
- No test evidence yet.
No score. We do not estimate one.
Weights and rules are on How we rank.