Multi-turn task completion on SearchGym
92.8Task Completion ScoreGemini-1.5-Pro
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-1.5-ProCategory=Closed-Source LLM2026.04 | 92.8 | |
| Gemini-1.5-FlashCategory=Closed-Source LLM2026.04 | 92.8 | |
| Qwen2.5-14B + PRIMECategory=PRIME Models2026.04 | 89.2 | |
| GRPO + Qwen2.5-8BCategory=RL (GRPO) Trained Models2026.04 | 88.8 | |
| GPT-4oCategory=Closed-Source LLM2026.04 | 88 | |
| Qwen2.5-8B + PRIMECategory=PRIME Models2026.04 | 87.2 | |
| GRPO + Qwen2.5-4BCategory=RL (GRPO) Trained Models2026.04 | 86.4 | |
| Qwen2.5-4B + PRIMECategory=PRIME Models2026.04 | 86 | |
| Qwen2.5-4BCategory=Open-Source LLM2026.04 | 85.6 | |
| Qwen2.5-8BCategory=Open-Source LLM2026.04 | 80.8 | |
| Qwen2.5-32BCategory=Open-Source LLM2026.04 | 79.2 | |
| Qwen2.5-14BCategory=Open-Source LLM2026.04 | 51.2 | |
| GPT-4o-miniCategory=Closed-Source LLM2026.04 | 35.2 |