Ko-AgentBench - L4 Parallel Tool Reasoning: leaderboard
Ko-AgentBench level 4 — Korean-language agent tasks whose independent tool calls can be planned and executed in parallel; accuracy.
Metric: Coverage (%). Source: huggingface.co. Saturation forecast: Not forecast. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 75 | #142 |
| 2 | Claude Haiku 4.5 | 75 | #280 |
| 3 | GPT-4o | 71.67 | #343 |
| 4 | Qwen 3 Next 80B A3B | 70 | #320 |
| 5 | DeepSeek V3.1 | 66.67 | #264 |
| 6 | Nova 2.0 Lite (Non-reasoning) | 65 | #417 (Nova 2.0 Lite) |
| 7 | GPT-4o Mini | 58.33 | #599 |
| 8 | GLM-4.6V | 56.67 | #316 |
| 9 | Grok 4.1 Fast | 56.67 | #217 |
| 10 | Nova Lite | 56.67 | #754 |
| 11 | GPT-5 | 55 | #98 |
| 12 | Gemini 2.5 Pro | 46.67 | #152 |
| 13 | GPT-5 Mini | 36.67 | #186 |
| 14 | Gemini 2.5 Flash | 35 | #245 |
| 15 | Gemini 2.5 Flash Lite | 35 | #422 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ko-agentbench-l4-parallel-tool-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.