Round 2: LLM To-Do App Battle Royale - 13 Models Tested!
🚀 The LLM Arena Expands: Round 2 Results
After our previous battle of 7 LLMs, the AI landscape has evolved dramatically! We’re back with 13 models - including returning champions and exciting newcomers. The same To-Do app challenge, but with fresh competition and updated models. Let’s see who claims the crown! 👑
📊 Complete Leaderboard
| Rank | LLM | Speed | Cost | Quality | Overall |
|---|---|---|---|---|---|
| 🥇 | Gemini flash 2.5-0520 | ★★★★★★★★★★ 10/10 | ★★★★★★★★★☆ 9/10 | ★★★★★★★★★☆ 9/10 | 9.3 |
| 🥈 | Devstrall Small 2505 | ★★★★★★★★★★ 10/10 | ★★★★★★★★★★ 10/10 | ★★★★★★☆☆☆☆ 6/10 | 8.7 |
| 🥉 | Gemini flash 2.5 experimental | ★★★★★★★★★★ 10/10 | ★★★★★★★★★★ 10/10 | ★★★★★★★★★☆ 9/10 | 9.7 |
| 4 | Gemini flash 2.0 | ★★★★★★★★★★ 10/10 | ★★★★★★★★★★ 10/10 | ★★★★★★★★☆☆ 8/10 | 9.3 |
| 5 | LLama 3.3 | ★★★★★★★★★★ 10/10 | ★★★★★★★★★★ 10/10 | ★★★★★★☆☆☆☆ 6/10 | 8.7 |
| 6 | DeepSeek 3.7 0324 | ★★★★★★☆☆☆☆ 6/10 | ★★★★★★★☆☆☆ 7/10 | ★★★★★★★★☆☆ 8/10 | 7.0 |
| 7 | OpenAI GPT 4.1 | ★★★★★★★☆☆☆ 7/10 | ★★★★★★☆☆☆☆ 6/10 | ★★★★★★★★★☆ 9/10 | 7.3 |
| 8 | Sonnet 4.0 | ★★★★★★☆☆☆☆ 6/10 | ★★☆☆☆☆☆☆☆☆ 2/10 | ★★★★★★★☆☆☆ 7/10 | 5.0 |
| 9 | Llama 4 Maverick | ★★★★☆☆☆☆☆☆ 4/10 | ★★★★★★★★★☆ 9/10 | ★★★★★★☆☆☆☆ 6/10 | 6.3 |
| 10 | Mistral Large 24-11 | ★★★★★★☆☆☆☆ 6/10 | ★★★★★★★★★★ 10/10 | ★★★★★☆☆☆☆☆ 5/10 | 7.0 |
| 11 | DeepSeek R1 | ★★★☆☆☆☆☆☆☆ 3/10 | ★★★★★★☆☆☆☆ 6/10 | ★★★★★★★☆☆☆ 7/10 | 5.3 |
| 12 | Claude Sonnet 3.7 | ★★★★★☆☆☆☆☆ 5/10 | ★★☆☆☆☆☆☆☆☆ 2/10 | ★★★★★★★★☆☆ 8/10 | 5.0 |
| 13 | Gemini Pro Preview 05-06 | ★★★★★★★☆☆☆ 7/10 | ★★☆☆☆☆☆☆☆☆ 2/10 | ★★★★★★★★☆☆ 8/10 | 5.7 |
🔍 Detailed Model Reviews
🥇 Gemini flash 2.5-0520 - The New Champion!
- Speed: 10/10 (215 tokens/s) ★★★★★★★★★★
- Cost: 9/10 ($0.009) ★★★★★★★★★☆
- Quality: 9/10 ★★★★★★★★★☆
🎯 The Perfect Balance: This model delivers exceptional performance across all metrics! Lightning-fast generation, reasonable cost, and outstanding code quality with comprehensive documentation and modern JavaScript practices. The drag-and-drop works flawlessly, and the dark mode implementation is top-notch. Minor deviations from prompt requirements prevent a perfect score.