Ko-AgentBench - L3 Sequential Tool Reasoning: leaderboard
Ko-AgentBench level 3 — Korean-language agent tasks requiring a dependent sequence of tool calls; accuracy.
Metric: PSM (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 100 | #343 |
| 2 | Claude Sonnet 4.5 | 100 | #142 |
| 3 | Claude Haiku 4.5 | 100 | #280 |
| 4 | Grok 4.1 Fast | 100 | #217 |
| 5 | Nova 2.0 Lite (Non-reasoning) | 100 | #417 (Nova 2.0 Lite) |
| 6 | Qwen 3 Next 80B A3B | 100 | #320 |
| 7 | GLM-4.6V | 96.67 | #316 |
| 8 | GPT-4o Mini | 91.67 | #599 |
| 9 | Nova Lite | 85 | #754 |
| 10 | DeepSeek V3.1 | 80 | #264 |
| 11 | GPT-5 | 56.67 | #98 |
| 12 | Gemini 2.5 Flash | 55 | #245 |
| 13 | GPT-5 Mini | 55 | #186 |
| 14 | Gemini 2.5 Pro | 26.67 | #152 |
| 15 | Gemini 2.5 Flash Lite | 20 | #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-l3-sequential-tool-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.