List your company

Long-context language model evaluation on HELMET

55.2Average ScoreFullKV
27.827234.933642.0449.1464Mar 12, 2026Mar 16, 2026Mar 21, 2026Mar 26, 2026Mar 31, 2026Apr 5, 2026Apr 10, 2026

Evaluation results

MethodLinks
FullKVLLM=Qwen2.5-7B-Instruct, KV Cache Size=Full2026.0355.2747153.875.631.820.5330.335649.8357.6786.67———————
DapQLLM=Qwen2.5-7B-Instruct, KV Cache Size=20482026.0354.787167.8557633.619.0529.4656.3348.8358.587———————
SnapKVLLM=Qwen2.5-7B-Instruct, KV Cache Size=20482026.0353.7568.46251.47833.220.3529.4155.6747.8358.3386.67———————
PyramidKVLLM=Qwen2.5-7B-Instruct, KV Cache Size=20482026.0353.5868.461.453.277.433.219.4628.5253.674860.8385.33———————
RecaLLM-Qwen2.5-7B2026.0453.5———————————96.269.469.713.746.235.544
RecaLLM-Llama-3.1-8B2026.0453.4———————————98.268.164.124.753.234.831
H2OLLM=Qwen2.5-7B-Instruct, KV Cache Size=20482026.0352.2463.2545179.431.420.3728.745249.3358.1787———————
DapQLLM=Qwen2.5-7B-Instruct, KV Cache Size=10242026.0352.076656.449.876.231.220.228.7654.3346.6757.585.67———————
SnapKVLLM=Qwen2.5-7B-Instruct, KV Cache Size=10242026.034956.842.1387530.820.5328.05564858.3385.37———————
PyramidKVLLM=Qwen2.5-7B-Instruct, KV Cache Size=10242026.0348.5958.440.63978.229.220.328.1150.6744.56184.5———————
DapQLLM=Qwen2.5-7B-Instruct, KV Cache Size=5122026.0348.15243.240.872.228.420.5429.55445.3357.3385.83———————
FullKVLLM=Llama3-8B-Instruct, KV Cache Size=Full2026.0345.7538.673.676.239.4251216.565242.1747.6780———————
H2OLLM=Qwen2.5-7B-Instruct, KV Cache Size=10242026.0345.3744.633.22879.425.419.5227.2450.6748.55785.5———————
SnapKVLLM=Qwen2.5-7B-Instruct, KV Cache Size=5122026.0343.7444.623.622.271.826.820.1727.855347.1758.1785.83———————
DapQLLM=Llama3-8B-Instruct, KV Cache Size=10242026.0342.9726.864.472.639.815.411.8116.67504548.3381.83———————
H2OLLM=Llama3-8B-Instruct, KV Cache Size=10242026.0342.85326470.63621.211.315.2847.6744.1747.6781.5———————
DapQLLM=Qwen2.5-7B-Instruct, KV Cache Size=2562026.0342.6841.425.427.667.42117.6426.5551.33475985.17———————
PyramidKVLLM=Qwen2.5-7B-Instruct, KV Cache Size=5122026.0342.49452022.870.625.821.9126.648.6743.559.6782.83———————
SnapKVLLM=Llama3-8B-Instruct, KV Cache Size=10242026.0340.892354.470.62918.211.0417.0752.6743.8348.3381.67———————
PyramidKVLLM=Llama3-8B-Instruct, KV Cache Size=10242026.0340.7324.855.468.629.61611.611652.3344.1748.1781.33———————
DapQLLM=Llama3-8B-Instruct, KV Cache Size=5122026.0340.5821.652.86439.215.411.1816.89494648.1782.17———————
H2OLLM=Qwen2.5-7B-Instruct, KV Cache Size=5122026.0340.3628.619.41776.217.818.5126.8851.2745.3357.6785.33———————
SnapKVLLM=Qwen2.5-7B-Instruct, KV Cache Size=2562026.0339.7338.415.218.86019.819.926.555145.3359.8382.17———————
H2OLLM=Llama3-8B-Instruct, KV Cache Size=5122026.0338.841851.86328.81810.6514.4548.3344.174882———————
SnapKVLLM=Llama3-8B-Instruct, KV Cache Size=5122026.0337.4717.641.463.622.214.211.3416.6750.6744.547.1782.83———————
Qwen2.5-7B-Instruct2026.0437.4———————————58.858.641.710.317.733.241.3
H2OLLM=Qwen2.5-7B-Instruct, KV Cache Size=2562026.0337.221.2129.274.414.418.2626.1248.674358.6783.33———————
ProLong-8B-Instruct-512KContext Length=512K2026.0437———————————74.368.718.612.412.536.436.5
DapQLLM=Llama3-8B-Instruct, KV Cache Size=2562026.0336.691737.44738.413.811.6615.748.674448.1781.83———————
PyramidKVLLM=Llama3-8B-Instruct, KV Cache Size=5122026.0336.3817.439.85622.212.211.6915.8750.6744.174882.17———————
Llama-3.1-8B-Instruct2026.0435.7———————————776331021.336.339.2
PyramidKVLLM=Qwen2.5-7B-Instruct, KV Cache Size=2562026.0335.4934.81413.250.81517.3625.93454059.1775.17———————
H2OLLM=Llama3-8B-Instruct, KV Cache Size=2562026.0334.2311.635.454.223.411.211.2112.8246.6742.334780.67———————
SnapKVLLM=Llama3-8B-Instruct, KV Cache Size=2562026.0333.3716.827.847.617.8911.5214.7650.6742.6746.8381.67———————
DapQLLM=Llama3-8B-Instruct, KV Cache Size=1282026.0332.616.423.425.63113.210.9915.1449.3341.547.3384.67———————
PyramidKVLLM=Llama3-8B-Instruct, KV Cache Size=2562026.0332.261427.436.2171011.3614.945242.834881.17———————
H2OLLM=Llama3-8B-Instruct, KV Cache Size=1282026.0329.166.82032.8186.210.7612.5245.33414681.33———————
SnapKVLLM=Llama3-8B-Instruct, KV Cache Size=1282026.032913.819.621.414.68.810.9213.664638.8347.0784.33———————
PyramidKVLLM=Llama3-8B-Instruct, KV Cache Size=1282026.0328.8811.221.819.614.88.210.714.034839.3346.583.5———————
LoongRL-14B2026.04—————————————63.7—18.7——49.8
LoongRL-7B2026.04—————————————61.6—11.4——44.5
QwenLong-L1-32BContext Length=32K2026.04—————————————52.9—14.3——47