Ko-AgentBench - L7 Long-Context Memory: leaderboard
Ko-AgentBench level 7 — Korean-language agent tasks requiring tool use while retaining information across a long context; accuracy.
Metric: Context Retention (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 100 | #98 |
| 2 | Claude Sonnet 4.5 | 100 | #142 |
| 3 | Claude Haiku 4.5 | 100 | #280 |
| 4 | GPT-5 Mini | 100 | #186 |
| 5 | DeepSeek V3.1 | 100 | #264 |
| 6 | Gemini 2.5 Flash Lite | 97.5 | #422 |
| 7 | Grok 4.1 Fast | 97.5 | #217 |
| 8 | Qwen 3 Next 80B A3B | 97.5 | #320 |
| 9 | GPT-4o | 95 | #343 |
| 10 | Nova 2.0 Lite (Non-reasoning) | 95 | #417 (Nova 2.0 Lite) |
| 11 | GPT-4o Mini | 92.5 | #599 |
| 12 | Gemini 2.5 Flash | 90 | #245 |
| 13 | GLM-4.6V | 90 | #316 |
| 14 | Gemini 2.5 Pro | 87.5 | #152 |
| 15 | Nova Lite | 72.5 | #754 |
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-l7-long-context-memory · How It Works · Data refreshed daily, snapshot 2026-10-20.