All Coding HarnessesVerified: 2026-11-19
Open-Source VS Code Extension (Cline Fork)24.2kv3.54.0Apache-2.0

Roo Code: 실제 토큰 배수 측정 및 BYOK 비용 절감 가이드

Multi-mode VS Code agent (Architect, Code, Debug, Ask) with per-mode system prompt pruning and model profiles.

code --install-extension RooVeterinaryInc.roo-cline
Install Friction
★☆☆☆☆
1/5 · Easy Setup
Runtime Dependency
VS Code >= 1.90Node.js >= 18 (optional for MCP)
BYOK & BaseURL
Full Support1 Env Vars
Typical Multiplier
1.7x – 2.6xTypical 2.0x vs Raw API

Context & Token Overhead Mechanics

In a 15-turn refactor with 5 tool calls, stripping unused tool schemas in Architect/Ask modes trims the multiplier to 1.7x–2.6x (~15% cheaper than Cline).

15-Turn Benchmark ≈ 61,500 Input Tokens
Base Task Payload:30,000 (49%)
System Prompt Overhead:4,700 (8%)
Context / Repo-Map:12,060 (20%)
Multi-Turn Tool Loop:14,740 (23%)
System Prompt Overhead
~2,350 tokens

Baseline instructions & tool schema weight per turn

Context Compression

Per-mode stripped system prompts (Ask/Architect exclude edit tools) + sliding window condense

Prompt Caching Hit Rate78%–90%

Separating Ask/Architect from Code trims base prompt overhead by ~25% vs single-mode extensions while preserving cache stability.

Model Compatibility Matrix & 15-Turn Cost

Estimated on 30k raw input + 4.5k output scaled by 2.0x multiplier at 70% cache hit rate

ModelTool CallingPrompt CachingVisionStreamingEst. 15-Turn Cost
DeepSeek: DeepSeek V4.1 FlashCode Mode BudgetNativeFullSupportedStable$0.002
Qwen: Qwen3 Coder 480B A35BSpecialized Open CoderNativeFullN/AStable$0.016
Anthropic: Claude Sonnet 5.5Orchestrator / ArchitectNativeFullSupportedStable$0.105
Google: Gemini 4 Neon1M Context ArchitectNativeFullSupportedStable$0.158

Production BYOK & DeepSeek / OpenRouter Config

Supports multiple saved API Configuration Profiles and binds different models + Base URLs to Architect, Code, Ask, and Orchestrator modes.

{
  "Code": {
    "apiProvider": "openai",
    "openAiBaseUrl": "https://api.deepseek.com/v1",
    "openAiApiKey": "${DEEPSEEK_API_KEY}",
    "openAiModelId": "deepseek-chat"
  },
  "Architect": {
    "apiProvider": "openrouter",
    "openRouterApiKey": "${OPENROUTER_API_KEY}",
    "openRouterModelId": "google/gemini-4-neon"
  }
}
Supported Env Vars:Roo Code API Profiles (Per-mode OpenRouter / DeepSeek / OpenAI-Compatible / Anthropic)

Roo Code Cost Pitfalls & Tuning FAQ

Each delegated subtask starts a fresh child context with its own 2.3k system prompt. Use Orchestrator only for multi-module work; keep small fixes in `Code` mode.

Compare Other Coding Harnesses

All 7 Benchmarks