Open-domain long-horizon evaluation on BrowseComp-Plus (BCP) (test)
10.6AccuracyHiMPO
Evaluation Results
| Method | Links | |
|---|---|---|
| HiMPOContext budget=8k ctx., Compression strategy=mem_aware2026.06 | 10.6 | |
| HiMPOContext budget=8k ctx., Compression strategy=naive_recency2026.06 | 10.36 | |
| HiMPOContext budget=16k ctx., Compression strategy=mem_aware2026.06 | 9.76 | |
| HiMPOContext budget=4k ctx., Compression strategy=mem_aware2026.06 | 9.4 | |
| HiMPOContext budget=16k ctx., Compression strategy=naive_recency2026.06 | 9.28 | |
| HiMPOContext budget=full ctx.2026.06 | 9.16 | |
| SUPOContext budget=8k ctx., Compression strategy=mem_aware2026.06 | 8.19 | |
| SUPOContext budget=8k ctx., Compression strategy=naive_recency2026.06 | 8.07 | |
| MEMPOContext budget=8k ctx., Compression strategy=mem_aware2026.06 | 7.95 | |
| MEMPOContext budget=8k ctx., Compression strategy=naive_recency2026.06 | 7.71 | |
| SFT onlyContext budget=8k ctx., Compression strategy=mem_aware2026.06 | 7.59 | |
| ReActContext budget=4k ctx., Compression strategy=mem_aware2026.06 | 7.47 | |
| SFT onlyContext budget=8k ctx., Compression strategy=naive_recency2026.06 | 7.47 | |
| SUPOContext budget=16k ctx., Compression strategy=mem_aware2026.06 | 7.35 | |
| SUPOContext budget=4k ctx., Compression strategy=mem_aware2026.06 | 7.23 | |
| SUPOContext budget=16k ctx., Compression strategy=naive_recency2026.06 | 7.11 | |
| MEMPOContext budget=4k ctx., Compression strategy=mem_aware2026.06 | 6.99 | |
| MEMPOContext budget=16k ctx., Compression strategy=mem_aware2026.06 | 6.99 | |
| SUPOContext budget=full ctx.2026.06 | 6.99 | |
| SFT onlyContext budget=4k ctx., Compression strategy=mem_aware2026.06 | 6.75 | |
| SFT onlyContext budget=16k ctx., Compression strategy=mem_aware2026.06 | 6.75 | |
| MEMPOContext budget=16k ctx., Compression strategy=naive_recency2026.06 | 6.75 | |
| MEMPOContext budget=full ctx.2026.06 | 6.63 | |
| SFT onlyContext budget=16k ctx., Compression strategy=naive_recency2026.06 | 6.51 | |
| ReActContext budget=16k ctx., Compression strategy=naive_recency2026.06 | 6.39 | |
| ReActContext budget=8k ctx., Compression strategy=mem_aware2026.06 | 6.27 | |
| SFT onlyContext budget=full ctx.2026.06 | 6.27 | |
| ReActContext budget=8k ctx., Compression strategy=naive_recency2026.06 | 5.3 | |
| ReActContext budget=16k ctx., Compression strategy=mem_aware2026.06 | 5.3 | |
| ReActContext budget=4k ctx., Compression strategy=naive_recency2026.06 | 5.18 | |
| ReActContext budget=full ctx.2026.06 | 5.06 | |
| HiMPOContext budget=4k ctx., Compression strategy=naive_recency2026.06 | 4.7 | |
| SUPOContext budget=4k ctx., Compression strategy=naive_recency2026.06 | 2.53 | |
| MEMPOContext budget=4k ctx., Compression strategy=naive_recency2026.06 | 2.29 | |
| SFT onlyContext budget=4k ctx., Compression strategy=naive_recency2026.06 | 2.05 |