List your company

Long-context evaluation on RULER 32k

90.06Overall ScoreRTPurbo
2.85625.495548.13570.7745Feb 3, 2026Feb 20, 2026Mar 9, 2026Mar 26, 2026Apr 12, 2026Apr 29, 2026May 16, 2026

Evaluation results

MethodLinks
RTPurboselection_mode=top-p2026.0590.06————————97.8485.4491.33100—————64.672.6399.9599.1599.6
Full Attn2026.0589.65————————9881.891.4799.93—————64.672.6399.9598.8599.66
HySparseModel Size=7B2026.0289.310010099.898.299.676.288.894.791.238.895.1———————————
RazorAttn2026.0588.69————————92.0482.590.0799.8—————6571.8899.697.6599.66
Full-AttnModel Size=7B2026.0288.21001001009899.475.896.498.390.516.695.1———————————
HySparseModel Size=80B2026.0287.410010099.496.89998.489.585.789.620.882.1———————————
RTPurboselection_mode=top-k2026.0584.36————————97.6882.59194.06—————58.471.7899.798.5565.53
Hybrid SWAModel Size=7B2026.0284.210010099.696.298.853.493.487.298.510.488.5———————————
Minference2026.0583.58————————95.8480.8885.3398.8—————6269.8897.4591.3570.73
SnapKV2026.0583.43————————94.2376.2389.9299.86—————58.9269.29471.197.46
FlexPrefill2026.0583.4————————96.5675.2690100—————60.471.1898.260.398.66
Full-AttnModel Size=80B2026.0282.110099.281.29999.4778679.574.540.766.7———————————
Quest2026.0578.97————————97.4857.2682.690.27—————56.868.7385.685.286.8
QUOKABackbone=Qwen-3-4B, BSA (Budget Size of Attention)=1,0242026.0274.83——————————————————————
Hybrid SWAModel Size=80B2026.0269.510098.258.889.690.26174.156.673.18.454.3———————————
QUOKABackbone=Smollm3, BSA (Budget Size of Attention)=1,0242026.0261.37——————————————————————
QUOKABackbone=Qwen-2.5-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0259.37——————————————————————
QUOKABackbone=GPT-OSS-20B, BSA (Budget Size of Attention)=1,0242026.0257.79——————————————————————
QUOKABackbone=Llama-3.2-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0257.01——————————————————————
SampleAttnBackbone=Smollm3, BSA (Budget Size of Attention)=1,0242026.0245.98——————————————————————
SampleAttnBackbone=Qwen-3-4B, BSA (Budget Size of Attention)=1,0242026.0240.72——————————————————————
LokiBackbone=GPT-OSS-20B, BSA (Budget Size of Attention)=1,0242026.0239.92——————————————————————
LokiBackbone=Qwen-3-4B, BSA (Budget Size of Attention)=1,0242026.0239.31——————————————————————
SparQBackbone=Qwen-2.5-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0236.74——————————————————————
SampleAttnBackbone=Qwen-2.5-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0236.17——————————————————————
SparQBackbone=Qwen-3-4B, BSA (Budget Size of Attention)=1,0242026.0235.2——————————————————————
LokiBackbone=Qwen-2.5-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0234.12——————————————————————
SampleAttnBackbone=Llama-3.2-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0231.73——————————————————————
SparQBackbone=Llama-3.2-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0231.14——————————————————————
SampleAttnBackbone=GPT-OSS-20B, BSA (Budget Size of Attention)=1,0242026.0230.42——————————————————————
KeyDiffBackbone=Smollm3, BSA (Budget Size of Attention)=1,0242026.0230.37——————————————————————
LessIsMoreBackbone=Smollm3, BSA (Budget Size of Attention)=1,0242026.0224.21——————————————————————
LokiBackbone=Smollm3, BSA (Budget Size of Attention)=1,0242026.0222.66——————————————————————
LessIsMoreBackbone=GPT-OSS-20B, BSA (Budget Size of Attention)=1,0242026.0220.11——————————————————————
LessIsMoreBackbone=Llama-3.2-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0219.16——————————————————————
SparQBackbone=Smollm3, BSA (Budget Size of Attention)=1,0242026.0218.69——————————————————————
SnapKVBackbone=GPT-OSS-20B, BSA (Budget Size of Attention)=1,0242026.0216.63——————————————————————
SparQBackbone=GPT-OSS-20B, BSA (Budget Size of Attention)=1,0242026.0215.2——————————————————————
KeyDiffBackbone=Llama-3.2-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0214.87——————————————————————
LessIsMoreBackbone=Qwen-3-4B, BSA (Budget Size of Attention)=1,0242026.0214.87——————————————————————
SnapKVBackbone=Smollm3, BSA (Budget Size of Attention)=1,0242026.0212.23——————————————————————
KeyDiffBackbone=Qwen-3-4B, BSA (Budget Size of Attention)=1,0242026.0212.07——————————————————————
KeyDiffBackbone=Qwen-2.5-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0210.32——————————————————————
LessIsMoreBackbone=Qwen-2.5-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.0210.12——————————————————————
SnapKVBackbone=Qwen-3-4B, BSA (Budget Size of Attention)=1,0242026.0210.07——————————————————————
KeyDiffBackbone=GPT-OSS-20B, BSA (Budget Size of Attention)=1,0242026.029.65——————————————————————
SnapKVBackbone=Qwen-2.5-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.028.13——————————————————————
LokiBackbone=Llama-3.2-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.028.05——————————————————————
SnapKVBackbone=Llama-3.2-3B-Instruct, BSA (Budget Size of Attention)=1,0242026.026.21——————————————————————
Engram-27BPre-training Steps=41k, Pre-training Loss=1.66, Context Window Size=32k2026.01—————————83.23.899.699.688.39389.544——————
Engram-27BPre-training Steps=46k, Pre-training Loss=1.63, Context Window Size=32k2026.01—————————87.24.398.697.68995.59737.5——————
Engram-27BPre-training Steps=50k, Pre-training Loss=1.62, Context Window Size=32k2026.01—————————895.999.399.389.396.59740.5——————
Gemma3-4BModel Size=4B2026.04—————————————————43.44—————
Llama3.2-3BModel Size=3B2026.04—————————————————51.38—————
MoE-27BPre-training Steps=50k, Pre-training Loss=1.63, Context Window Size=32k2026.01—————————774.5731008892.784.234.5——————
Qwen2.5-3BModel Size=3B2026.04—————————————————58.95—————
Qwen3-4B-BaseModel Size=4B, Training Strategy=Base2026.04—————————————————62.28—————
Qwen3-4B-Instruct-2507Model Size=4B, Training Strategy=Instruct-25072026.04—————————————————85.62—————
SpB1.0-7BModel Size=7B2026.04—————————————————17.21—————
SpB2.0-5BModel Size=5B2026.04—————————————————50.87—————