List your company

Character description generation on BookWorm

22.05PRISMAGPT-4.1 mini (Full context)
11.566814.288417.0119.7316Apr 13, 2026

Evaluation results

MethodLinks
GPT-4.1 mini (Full context)Model=GPT-4.1 mini, Context=Full context, Trace=none, Protocol=Zero-Shot2026.0422.0516.6775.5335.4117.71
Qwen-3-8b (Hierarchical-16k)Model=Qwen-3-8b, Context=Hierarchical-16k, Trace=none, Protocol=Zero-Shot2026.0419.9915.7970.5736.5618.22
Qwen-3-8b (Hierarchical-16k + guided-QA)Model=Qwen-3-8b, Context=Hierarchical-16k, Trace=ZS, Protocol=Zero-Shot2026.0419.8216.270.7437.6717.9
Qwen-3-8b (Hierarchical-16k + reasoning)Model=Qwen-3-8b, Context=Hierarchical-16k, Trace=built-in, Protocol=Zero-Shot2026.0419.7415.1671.1535.8217.33
Qwen-3-8b (Coref-32k + guided-QA)Model=Qwen-3-8b, Context=Coref-32k, Trace=GRPO, Protocol=Zero-Shot2026.0419.4914.8651.2337.6618.01
Qwen-3-8b (Coref-32k + guided-QA)Model=Qwen-3-8b, Context=Coref-32k, Trace=GRPO, Protocol=SFT2026.0419.3215.15237.5818.42
Qwen-3-8b (Lead-128k w/ YaRN + guided-QA)Model=Qwen-3-8b, Context=Lead-128k (w/ YaRN), Trace=ZS, Protocol=Zero-Shot2026.0419.3215.7771.5537.817.77
Qwen-3-8b (Lead-128k w/ YaRN + guided-QA)Model=Qwen-3-8b, Context=Lead-128k (w/ YaRN), Trace=GRPO, Protocol=Zero-Shot2026.0419.316.2270.5137.1218.39
Qwen-3-8b (Coref-32k)Model=Qwen-3-8b, Context=Coref-32k, Trace=none, Protocol=Zero-Shot2026.0419.1113.9652.5836.1218.44
Qwen-3-8b (Hierarchical-16k + guided-QA)Model=Qwen-3-8b, Context=Hierarchical-16k, Trace=GRPO, Protocol=Zero-Shot2026.0418.9415.3667.9236.4918.19
Qwen-3-8b (Coref-32k + guided-QA)Model=Qwen-3-8b, Context=Coref-32k, Trace=ZS, Protocol=Zero-Shot2026.0418.6315.0150.636.9517.5
Qwen-3-8b (Coref-32k + reasoning)Model=Qwen-3-8b, Context=Coref-32k, Trace=built-in, Protocol=Zero-Shot2026.0418.5913.9449.9435.8816.54
Qwen-3-8b (BM25-32k + guided-QA)Model=Qwen-3-8b, Context=BM25-32k, Trace=GRPO, Protocol=Zero-Shot2026.0418.3815.1859.937.0517.59
Qwen-3-8b (BM25-32k + guided-QA)Model=Qwen-3-8b, Context=BM25-32k, Trace=GRPO, Protocol=SFT2026.0418.2715.1560.4636.617.87
Qwen-3-8b (BM25-32k + guided-QA)Model=Qwen-3-8b, Context=BM25-32k, Trace=ZS, Protocol=Zero-Shot2026.0417.8414.6659.9636.2317.43
Qwen-3-8b (Incremental-16k)Model=Qwen-3-8b, Context=Incremental-16k, Trace=none, Protocol=Zero-Shot2026.0417.6414.4269.6835.5917.58
Qwen-3-8b (Coref-32k)Model=Qwen-3-8b, Context=Coref-32k, Trace=none, Protocol=SFT2026.0417.4614.3652.5435.5919.26
GPT-4.1 mini (No context)Model=GPT-4.1 mini, Context=No context, Trace=none, Protocol=Zero-Shot2026.0417.3612.49—29.3517.41
Qwen-3-8b (BM25-32k)Model=Qwen-3-8b, Context=BM25-32k, Trace=none, Protocol=Zero-Shot2026.0417.0913.5658.5834.0618.18
Qwen-3-8b (BM25-32k + reasoning)Model=Qwen-3-8b, Context=BM25-32k, Trace=built-in, Protocol=Zero-Shot2026.0417.0613.558.4435.1316.51
Qwen-3-8b (Lead-128k w/ YaRN)Model=Qwen-3-8b, Context=Lead-128k (w/ YaRN), Trace=none, Protocol=Zero-Shot2026.0417.0213.6869.2934.9418.11
Qwen-3-8b (Lead-128k w/ YaRN + reasoning)Model=Qwen-3-8b, Context=Lead-128k (w/ YaRN), Trace=built-in, Protocol=Zero-Shot2026.0416.9313.6567.1734.216.43
Qwen-3-8b (Incremental-16k + guided-QA)Model=Qwen-3-8b, Context=Incremental-16k, Trace=ZS, Protocol=Zero-Shot2026.0416.814.5668.234.9216.98
Qwen-3-8b (Incremental-16k + guided-QA)Model=Qwen-3-8b, Context=Incremental-16k, Trace=GRPO, Protocol=Zero-Shot2026.0416.5913.6366.0534.5117.42
Qwen-3-8b (BM25-32k)Model=Qwen-3-8b, Context=BM25-32k, Trace=none, Protocol=SFT2026.0416.1713.266033.1118.99
Qwen-3-8b (Incremental-16k + reasoning)Model=Qwen-3-8b, Context=Incremental-16k, Trace=built-in, Protocol=Zero-Shot2026.0416.0313.3666.4633.3715.76
Qwen-3-8b (Lead-32k)Model=Qwen-3-8b, Context=Lead-32k, Trace=none, Protocol=Zero-Shot2026.0415.7212.3561.0130.617.79
Qwen-3-8b (No Context)Model=Qwen-3-8b, Context=No Context, Trace=none, Protocol=Zero-Shot2026.0411.9710.31—25.9717.55