Agent Cache Engineering - AKV - xbench-DS 2510: leaderboard
Agent Cache Engineering (2026): AKV context management, SGLang BF16 weights on A800 GPUs; GPT-OSS High and the other two models Thinking. GAIA is the103-task text validation subset; AKV retains12latest tool responses, Standard summarizes at96K/192K and Text Engineering at48K. Cost/latency replay metrics are separate. Success Rate (%).
Metric: Success Rate (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | MiroThinker-1.7-Mini BF16 (Thinking) | 43.5 | |
| 2 | MiniMax-M2.7 BF16 (Thinking) | 40.5 | |
| 3 | GPT-OSS-120B BF16 (High) | 37.5 |
Interactive version: theaggregate.ai/benchmark?slug=agent-cache-engineering-akv-xbench-ds-2510 · How It Works · Data refreshed daily, snapshot 2026-10-22.