List your company

Long-context language understanding suite on ZeroSCROLLS

33.5GovReport ScoreQwen3-4B
4.58812.09419.627.106Mar 5, 2024Jun 21, 2024Oct 7, 2024Jan 24, 2025May 12, 2025Aug 28, 2025Dec 15, 2025

Evaluation results

MethodLinks
Qwen3-4BInference Strategy=qTTT2025.1233.59.98.618.73438—32.530.587
Qwen3-8BInference Strategy=qTTT2025.1229.87.98.618.326.142.8—34.648.994.5
In-contextModel=Qwen3-32B, Test-time compute strategy=In-context2025.1226.7—11.9—24.127.7——28.9—
qTTTModel=Qwen3-32B, Test-time compute strategy=qTTT2025.1226—10.8—42.449.6——59.2—
Qwen3-1.7BInference Strategy=qTTT2025.12267.49.51831.111.7—25.826.276.2
Qwen3-4BInference Strategy=In-context2025.1224.99.910.99.923.211—18.417.140.5
ThinkingModel=Qwen3-32B, Test-time compute strategy=Thinking2025.1224.8—9.9—3542.4——54.9—
Qwen3-1.7BInference Strategy=In-context2025.1222.58.26.29.225.715—18.311.647.6
Qwen3-8BInference Strategy=In-context2025.122210.49.818.119.618.9—24.122.571.4
Qwen3-1.7BInference Strategy=Thinking2025.1221.87.27.514.621.48.9—20.722.661.9
Ms-PoEBackbone=Vicuna-7B-16K2024.0321.414.316.220.237.82143.324.9——
Ms-PoEBackbone=Vicuna-7B2024.0320.214.515.419.834.716.210.518.8——
BaselineBackbone=Vicuna-7B-16K2024.0320.213.916.220.132.318.829.921.6——
Qwen3-4BInference Strategy=Thinking2025.1220.28.37.716.824.730—23.97.576.2
Qwen3-8BInference Strategy=Thinking2025.1217.87.98.615.321.135.1—3043.990.5
Ms-PoEBackbone=Llama-2-7B-Chat2024.0317.714.215.819.925.117.75.816.6——
BaselineBackbone=Llama-2-7B-Chat2024.0316.814.115.219.521.914.43.115——
Ms-PoEBackbone=StableBeluga-7B2024.0316.614.215.218.736.91814.219.1——
Ms-PoEBackbone=Llama-2-13B-Chat2024.0316.513.115.519.220.8175.915.4——
BaselineBackbone=Vicuna-7B2024.0316.212.715.118.924.313.73.315——
BaselineBackbone=Llama-2-13B-Chat2024.0315.412.315.118.919155.714.5——
BaselineBackbone=StableBeluga-7B2024.0314.913.814.717.928.116.89.216.5——
Ms-PoEBackbone=StableBeluga-13B2024.037.47.412.813.220.813.45.611.5——
BaselineBackbone=StableBeluga-13B2024.035.77.112.913.319.213.44.810.9——