List your company

Long-context Understanding on LongBench (Full Suite Evaluation)

63.4Average ScoreLongCrafter
9.767223.691137.61551.5389Feb 6, 2025May 3, 2025Jul 28, 2025Oct 22, 2025Jan 16, 2026Apr 12, 2026Jul 7, 2026

Evaluation results

MethodLinks
LongCrafterBackbone=LLaMA-3.1-8B, SFT Method=LongCrafter2026.0763.4—53—72.570.557.5——————————
LongCrafterBackbone=Qwen2.5-7B, SFT Method=LongCrafter2026.0762.8—52.5—75.57449——————————
LongFaithBackbone=LLaMA-3.1-8B, SFT Method=LongFaith2026.0760.5—45.5—697156.5——————————
LongFaithBackbone=Qwen2.5-7B, SFT Method=LongFaith2026.0759.4—47—73.57146——————————
LongMagpieBackbone=LLaMA-3.1-8B, SFT Method=LongMagpie2026.0756.9—51.5—67.562.546——————————
Qwen2.5-7B-InstructBackbone=Qwen2.5-7B, SFT Method=Instruct (official)2026.0756.8—57.5—746035.5——————————
LLaMA-3.1-8B-InstructBackbone=LLaMA-3.1-8B, SFT Method=Instruct (official)2026.0753.25—53—695833——————————
LongMagpieBackbone=Qwen2.5-7B, SFT Method=LongMagpie2026.0752.9—50.5—675539——————————
LongRewardBackbone=Qwen2.5-7B, SFT Method=LongReward2026.0752.5—50.5—67.55438——————————
LongRewardBackbone=LLaMA-3.1-8B, SFT Method=LongReward2026.0749.4—50—63.545.538.5——————————
LongAlignBackbone=Qwen2.5-7B, SFT Method=LongAlign2026.0748.9—51—63.551.529.5——————————
FullBackbone=Mistral-7B-Instruct-v0.3, Cache Budget=Full Cache2026.0548.1129.0741.5452.8849.3739.0128.5835.0725.7127.737688.5947.51698.561.4862.68
HeadKV w/ CriticalKVBackbone=Llama-3.1-8B, Cache Budget=40%2025.0248————————————————
AdaKV w/ CriticalKVBackbone=Llama-3.1-8B, Cache Budget=40%2025.0247.79————————————————
SnapKV w/ CriticalKVBackbone=Llama-3.1-8B, Cache Budget=40%2025.0247.29————————————————
LongAlignBackbone=LLaMA-3.1-8B, SFT Method=LongAlign2026.0746.6—51—604530.5——————————
ReST-KVBackbone=Mistral-7B-Instruct-v0.3, Cache Budget=512L2026.0546.4728.635.8653.3749.1338.727.9426.0524.3725.0973.589.6646.275.598.560.1360.84
SnapKV w/ OBCache (Joint)Backbone=Llama-3.1-8B, Cache Budget=40%2025.0245.92————————————————
SnapKVBackbone=Mistral-7B-Instruct-v0.3, Cache Budget=512L2026.0545.6626.8435.5153.1249.5637.7226.5425.0624.0324.7667.589.3644.825.598.560.4461.22
SnapKVBackbone=Llama-3.1-8B, Cache Budget=40%2025.0245.52————————————————
TOVABackbone=Mistral-7B-Instruct-v0.3, Cache Budget=512L2026.0542.8525.2332.5246.2445.2336.2320.3224.5322.5323.6466.587.2444.21685.6259.3560.24
H2OBackbone=Mistral-7B-Instruct-v0.3, Cache Budget=512L2026.0541.4225.2330.4140.3242.5235.2318.2324.2321.2423.2166.586.7143.15582.5260.1358.15
ReST-KVBackbone=Mistral-7B-Instruct-v0.3, Cache Budget=64L2026.0540.0525.6526.5842.7146.1136.4324.3419.821.6518.951.587.8841.54490.552.3950.75
FullBackbone=Llama-3.1-8B-Instruct, Cache Budget=Full Cache2026.0539.9632.0213.1227.5216.616.4111.4134.5923.4126.897391.6543.87.1897.7365.1258.89
StreamingLLMBackbone=Mistral-7B-Instruct-v0.3, Cache Budget=512L2026.0539.6824.1925.9730.1440.7531.917.3522.1820.323.2265.586.9543.7568159.3556.36
ReST-KVBackbone=Llama-3.1-8B-Instruct, Cache Budget=512L2026.0538.3332.0110.7325.2315.9115.8510.2526.4723.2324.796991.6242.598.497.6663.4856.03
SnapKVBackbone=Mistral-7B-Instruct-v0.3, Cache Budget=64L2026.0537.5319.3923.6238.6643.2634.7221.3317.5920.9317.0638.586.9639.61790.551.6349.73
SnapKVBackbone=Llama-3.1-8B-Instruct, Cache Budget=512L2026.0537.4728.029.8324.8413.7715.410.2125.1322.7324.256592.3441.698.4296.3164.357.28
TOVABackbone=Mistral-7B-Instruct-v0.3, Cache Budget=64L2026.0536.522.5122.2437.2341.1234.119.5217.2119.2316.2738.585.1238.516.586.551.0448.42
TOVABackbone=Llama-3.1-8B-Instruct, Cache Budget=512L2026.0535.5927.348.3422.4512.2514.518.4224.2322.1322.2558.589.3140.518.2493.1462.2355.61
H2OBackbone=Mistral-7B-Instruct-v0.3, Cache Budget=64L2026.0534.9320.5121.5225.1240.1233.1218.3416.2319.1216.2438.583.1237.23685.550.1248.12
H2OBackbone=Llama-3.1-8B-Instruct, Cache Budget=512L2026.0534.5326.237.3420.5111.5213.527.3423.2321.2423.1458.586.1240.157.2591.0261.2354.12
StreamingLLMBackbone=Mistral-7B-Instruct-v0.3, Cache Budget=64L2026.0534.2420.3720.5624.6238.8732.4717.6815.4819.8415.8139.582.7736.725.58049.7747.9
StreamingLLMBackbone=Llama-3.1-8B-Instruct, Cache Budget=512L2026.0533.8419.156.4715.0210.9412.586.2323.6620.0523.3157.587.741.8610.2590.7462.3953.61
ReST-KVBackbone=Llama-3.1-8B-Instruct, Cache Budget=64L2026.0533.5422.437.1919.2514.1115.047.9720.5621.119.1553.588.2340.218.4693.956.7448.77
ReST-KVBackbone=Llama2-7B-Chat, Cache Budget=512L2026.0532.3817.1519.8832.7131.9425.629.9720.5220.6823.5963.583.339.29611.558.6553.81
SnapKVBackbone=Llama2-7B-Chat, Cache Budget=512L2026.0532.1816.2219.5732.3231.8724.979.6620.1920.7723.856282.2439.18610.559.4956.06
SnapKVBackbone=Llama-3.1-8B-Instruct, Cache Budget=64L2026.0531.4819.95.7818.3813.5114.428.5217.3520.4417.334185.3737.638.9391.0855.0948.88
TOVABackbone=Llama-3.1-8B-Instruct, Cache Budget=64L2026.0530.6318.526.1217.3212.1512.517.3516.2420.4116.3438.4182.6136.168.1495.2355.2147.35
TOVABackbone=Llama2-7B-Chat, Cache Budget=512L2026.0530.2413.5315.4626.4426.1231.027.1218.2518.6422.3462.583.140.613856.1451.53
StreamingLLMBackbone=Llama-3.1-8B-Instruct, Cache Budget=64L2026.0529.617.655.0814.1410.9312.646.8616.5718.9316.338.583.1334.659.7896.2854.1648.21
H2OBackbone=Llama-3.1-8B-Instruct, Cache Budget=64L2026.0529.5212.235.1215.1211.5110.146.2317.2319.5116.7939.1581.5136.128.1295.1251.2547.12
StreamingLLMBackbone=Llama2-7B-Chat, Cache Budget=512L2026.0528.8615.315.5320.1626.5925.055.6518.319.2821.8454.582.2338.075.5556.851.95
ReST-KVBackbone=Llama2-7B-Chat, Cache Budget=64L2026.0525.5412.7217.1724.0924.7123.85.5515.1819.7117.4543.576.1733.425.544540.61
SnapKVBackbone=Llama2-7B-Chat, Cache Budget=64L2026.0522.9610.8316.3817.5322.8123.245.0613.1218.3814.1734.569.4533.435.5739.9936.04
TOVABackbone=Llama2-7B-Chat, Cache Budget=64L2026.0520.28.2614.3412.6413.5213.253.5311.6416.6713.353672.6432.722436.1532.53
LaCacheBackbone=Llama2-7B-Chat, Cache Budget=64L2026.0513.518.6116.517.4215.1417.774.3613.0917.4613.8318.2518.1212.935.562417.24
H2OBackbone=Llama2-7B-Chat, Cache Budget=512L2026.0512.859.688.676.8610.858.711.3120.0418.7224.911817.0918.993.752.320.8714.87
H2OBackbone=Llama2-7B-Chat, Cache Budget=64L2026.0512.424.4612.148.8512.1113.342.3613.0616.6316.8919.520.6910.452.7326.516.06
StreamingLLMBackbone=Llama2-7B-Chat, Cache Budget=64L2026.0511.835.6115.516.4214.1416.771.3612.0916.4612.8317.2515.1210.934.532215.24