Roo Code is a VS Code extension for AI-assisted development with multiple modes. It supports various models and benefits from strong code generation and instruction following.
Best Models for Roo Code
Top 15 by tool-optimized score
Scored by: benchmark performance (90%) from MMLU, GPQA, HumanEval, SWE-bench, and 15+ standardized evaluations, with capabilities and context as tiebreakers (10%).
| # | Model | Score | Output $/M |
|---|---|---|---|
| 1 | Gemini 3.5 Flash Arena Elo: 1477 | 86 | $9.00 |
| 2 | Claude Fable 5 SWE-bench: 95% | 85 | $50.00 |
| 3 | Claude Opus 4.8 (Fast) | 85 | $50.00 |
| 4 | Claude Opus 4.7 (Fast) | 85 | $150.00 |
| 5 | MiMo-V2.5-Pro Arena Elo: 1466 | 85 | $0.870 |
| 6 | GLM 5.1 Arena Elo: 1475 | 85 | $3.08 |
| 7 | Qwen3.6 Max Preview Arena Elo: 1461 | 84 | $6.24 |
| 8 | Kimi K2.6 Arena Elo: 1460 | 84 | $3.41 |
| 9 | GLM 5 Arena Elo: 1457 | 84 | $1.92 |
| 10 | MiniMax M3 Arena Elo: 1448 | 83 | $1.20 |
| 11 | Grok 4.3 Arena Elo: 1446 | 83 | $2.50 |
| 12 | Claude Opus 4.6 (Fast) | 83 | $150.00 |
| 13 | Gemma 4 31B Arena Elo: 1451 | 83 | $0.350 |
| 14 | Qwen3.6 Plus Arena Elo: 1444 | 83 | $1.95 |
| 15 | GPT-5.4 Pro | 83 | $180.00 |
| 16 | Gemini 3.1 Pro Preview Custom Tools | 83 | $12.00 |
| 17 | Qwen3.5 397B A17B Arena Elo: 1444 | 83 | $2.45 |
| 18 | GPT-5.2-Codex | 83 | $14.00 |
| 19 | GLM 4.7 Arena Elo: 1443 | 83 | $1.75 |
| 20 | GPT-5.2 Pro | 83 | $168.00 |
| 21 | Claude Opus 4.1 Arena Elo: 1449 | 83 | $75.00 |
| 22 | Claude Opus 4.8 SWE-bench: 88.6% | 82 | $25.00 |
| 23 | GPT-5.5 Pro SWE-bench: 88.7% | 82 | $180.00 |
| 24 | GPT-5.5 SWE-bench: 88.7% | 82 | $30.00 |
| 25 | DeepSeek V4 Flash Arena Elo: 1433 | 82 | $0.180 |
| 26 | MiMo-V2.5 Arena Elo: 1433 | 82 | $0.280 |
| 27 | Gemma 4 26B A4B Arena Elo: 1438 | 82 | $0.330 |
| 28 | Grok 4.20 | 82 | $2.50 |
| 29 | Gemini 3.1 Flash Lite Preview Arena Elo: 1432 | 82 | $1.50 |
| 30 | GPT-5.3-Codex | 82 | $14.00 |
| 31 | GPT-5.2 Chat Arena Elo: 1475 | 82 | $14.00 |
| 32 | GPT-5 Pro | 82 | $120.00 |
| 33 | GPT-5 Codex | 82 | $10.00 |
| 34 | Mistral Medium 3.5 Arena Elo: 1426 | 81 | $7.50 |
| 35 | Claude Opus 4.7 SWE-bench: 87.6% | 81 | $25.00 |
| 36 | MiniMax M2.7 Arena Elo: 1417 | 81 | $0.960 |
| 37 | Qwen3.5-122B-A10B Arena Elo: 1417 | 81 | $2.08 |
| 38 | GPT-5.1-Codex-Max | 81 | $10.00 |
| 39 | GPT-5.1-Codex | 81 | $10.00 |
| 40 | GPT-5.1-Codex-Mini | 81 | $2.00 |
| 41 | o3 Deep Research | 81 | $40.00 |
| 42 | GLM 4.6 Arena Elo: 1425 | 81 | $1.74 |
| 43 | o3 Pro | 81 | $80.00 |
| 44 | Qwen3.5-27B Arena Elo: 1409 | 80 | $1.56 |
| 45 | Qwen3 VL 235B A22B Instruct Arena Elo: 1415 | 80 | $0.880 |
| 46 | Mistral Large HumanEval: 92% | 80 | $6.00 |
| 47 | Qwen3.5-35B-A3B Arena Elo: 1396 | 79 | $1.00 |
| 48 | Qwen3.5-Flash Arena Elo: 1397 | 79 | $0.260 |
| 49 | Claude Opus 4.6 SWE-bench: 83.7% | 79 | $25.00 |
| 50 | Step 3.5 Flash Arena Elo: 1395 | 79 | $0.300 |
Based on our analysis of coding benchmarks, capability matching, and pricing, Gemini 3.5 Flash currently ranks #1 for Roo Code. Rankings are rebuilt as benchmark, pricing, and provider data refresh.
We score models using benchmark performance (90%) from LMArena, HumanEval, SWE-bench, MMLU, and 15+ standardized evaluations. Capabilities and context serve as tiebreakers (10%). Only models with the capabilities Roo Code needs are included in the tool-specific rankings.
We currently track 316 AI models compatible with Roo Code. This includes models from OpenAI, Anthropic, Google, DeepSeek, and other providers accessible via API.
Many open-source models are compatible with Roo Code through API providers like OpenRouter, Together AI, and Groq. Check our rankings to see which open-source models perform best.
Rankings refresh whenever the underlying benchmark, pricing, and catalog sources refresh. That means some signals update faster than others, and the page reflects the latest verified source data available.