The best models for Roo Code
— free to start
Connect Roo Code to Kyma's active open models. Auto-failback, prompt caching, OpenAI-compatible.
Multi-Mode Ready
Different models for Architect, Code, Edit, and Debug modes. Kyma lets you switch models per mode without switching providers.
Auto-Failback
If one provider is slow, Kyma routes to the next. Your coding flow never stops.
Prompt Caching
Repeated context? Kyma caches it, and the repeated part bills at that model's own cached-input rate — listed beside its input rate on every model page.
Recommended Models for Roo Code
Picked for quality, speed, and tool use — matched to Roo Code's multi-mode workflow.
Top quality, 131K context. Perfect for Roo Code's Architect and Code modes.
qwen-3.6-plus#1 on Kyma
262K context, multimodal, strongest tool use. Ideal for autonomous coding.
kimi-k2.6Best Agentic
Highest SWE-bench globally. Use for complex refactoring.
minimax-m2.580.2% SWE-bench
Lightning fast inference. Best for Roo Code's quick edit mode.
qwen-3-32bUltra Fast
V4-tier quality at the lowest price. 1M context, native reasoning, MIT licensed.
deepseek-v4-flashBest value
Setup in 30 Seconds
1. Get your free API key → 2. Install Roo Code from the VS Code Marketplace → 3. Paste the config below.
1. Open VS Code → Extensions → Search "Roo Code" → Install 2. Click Roo Code icon → Settings 3. Set these values: API Provider: OpenAI Compatible Base URL: https://kymaapi.com/v1 API Key: ky-xxxxx (get yours at kymaapi.com) Model ID: qwen-3.6-plus Done! Roo Code now uses Kyma's models.
Roo Code supports multiple modes. Recommended models per mode: Architect Mode: deepseek-v4-pro (top reasoning, 1M context) Code Mode: kimi-k2.6 (best tool use) Edit Mode: qwen-3-32b (fastest) Debug Mode: deepseek-v4-flash (cheap deep analysis) Tip: Use model aliases for easy switching: "best" → qwen-3.6-plus "code" → qwen-3-coder "agent" → kimi-k2.6 "fast" → qwen-3-32b
Start coding with the best models — free
$0.50 free credits on signup. No credit card required.
Get Free API Key