List your company

Long-context Reasoning on LongBench

73.8ScoreGHG-TDA
30.74441.92253.164.278Dec 12, 2025Jan 14, 2026Feb 16, 2026Mar 22, 2026Apr 24, 2026May 27, 2026Jun 30, 2026

Evaluation results

MethodLinks
GHG-TDABase Model=GPT-4o2026.0273.8
GHG-TDABase Model=Claude 3.5 Sonnet2026.0273.5
AoTBase Model=GPT-4o2026.0272.6
AoTBase Model=Claude 3.5 Sonnet2026.0272.2
GHS-TDABackbone=DeepSeek-V32026.0269.9
GHS-TDABackbone=GPT-4o-mini2026.0269.5
AoTBackbone=DeepSeek-V3, Reasoning Paradigm=atomic reasoning2026.0268.7
GHS-TDABackbone=Qwen-Turbo2026.0268.6
AoTBackbone=GPT-4o-mini, Reasoning Paradigm=atomic reasoning2026.0268.5
AoTBackbone=Qwen-Turbo, Reasoning Paradigm=atomic reasoning2026.0268.1
GoTBase Model=Claude 3.5 Sonnet2026.0267
GoTBase Model=GPT-4o2026.0266.8
ToTBase Model=Claude 3.5 Sonnet2026.0266.4
ToTBase Model=GPT-4o2026.0266
GHS-TDABackbone=Llama 3-8B2026.0264.64
AoTBackbone=Llama 3-8B2026.0263.71
GoTBackbone=DeepSeek-V3, Reasoning Paradigm=graph-based2026.0263.4
ToTBackbone=DeepSeek-V3, Reasoning Paradigm=tree-based2026.0263.2
GoTBackbone=GPT-4o-mini, Reasoning Paradigm=graph-based2026.0263.1
GHS-TDABackbone=Qwen 2-14B2026.0262.8
ToTBackbone=GPT-4o-mini, Reasoning Paradigm=tree-based2026.0262.8
GoTBackbone=Qwen-Turbo, Reasoning Paradigm=graph-based2026.0262.8
ToTBackbone=Qwen-Turbo, Reasoning Paradigm=tree-based2026.0262.4
CoTBase Model=GPT-4o2026.0262
CoTBase Model=Claude 3.5 Sonnet2026.0261.7
AFlowBackbone=DeepSeek-V3, Reasoning Paradigm=search-based2026.0261.4
AoTBackbone=Qwen 2-14B2026.0261.2
AFlowBackbone=GPT-4o-mini, Reasoning Paradigm=search-based2026.0261
AFlowBackbone=Qwen-Turbo, Reasoning Paradigm=search-based2026.0260.4
ToTBackbone=Qwen 2-14B2026.0259.3
FoTBackbone=DeepSeek-V3, Reasoning Paradigm=forest-based, n=82026.0259.3
FoTBackbone=GPT-4o-mini, Reasoning Paradigm=forest-based, n=82026.0259.1
FoTBackbone=Qwen-Turbo, Reasoning Paradigm=forest-based, n=82026.0259
CoT-SCBackbone=DeepSeek-V3, Reasoning Paradigm=chain-based, n=52026.0258.9
GoTBackbone=Llama 3-8B2026.0258.68
CoT-SCBackbone=GPT-4o-mini, Reasoning Paradigm=chain-based, n=52026.0258.6
ToTBackbone=Llama 3-8B2026.0258.4
CoT-SCBackbone=Qwen-Turbo, Reasoning Paradigm=chain-based, n=52026.0258.4
Self-RefineBackbone=DeepSeek-V3, Reasoning Paradigm=chain-based2026.0258.4
Self-RefineBackbone=GPT-4o-mini, Reasoning Paradigm=chain-based2026.0258.2
GoTBackbone=Qwen 2-14B2026.0258.1
Self-RefineBackbone=Qwen-Turbo, Reasoning Paradigm=chain-based2026.0258
CoTBackbone=DeepSeek-V3, Reasoning Paradigm=chain-based2026.0257.7
CoTBackbone=GPT-4o-mini, Reasoning Paradigm=chain-based2026.0257.6
CoTBackbone=Qwen-Turbo, Reasoning Paradigm=chain-based2026.0257.3
Full KVBackbone=Qwen2.5-14B, KV Cache Budget=Full2026.0655.47
SEKVBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.0654.71
CoTBackbone=Llama 3-8B2026.0253.57
Analogical PromptingBackbone=DeepSeek-V3, Reasoning Paradigm=chain-based2026.0253.4
Analogical PromptingBackbone=GPT-4o-mini, Reasoning Paradigm=chain-based2026.0252.9
CoTBackbone=Qwen 2-14B2026.0252.7
Analogical PromptingBackbone=Qwen-Turbo, Reasoning Paradigm=chain-based2026.0252.7
SentenceKVBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.0652.31
SemantiCacheBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.0651.88
ChunkKVBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.0651.42
Full KVBackbone=Llama-3.1-8B, KV Cache Budget=Full2026.0651.18
SEKVBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.0651.02
Full KVBackbone=Llama-3-8B, KV Cache Budget=Full2026.0650.34
SEKVBackbone=Llama-3-8B, KV Cache Budget=10%2026.0650.18
PyramidKVBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.0650.17
SentenceKVBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.0649.93
SemantiCacheBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.0649.61
SnapKVBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.0649.34
ChunkKVBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.0649.14
SentenceKVBackbone=Llama-3-8B, KV Cache Budget=10%2026.0649.12
SemantiCacheBackbone=Llama-3-8B, KV Cache Budget=10%2026.0648.73
ChunkKVBackbone=Llama-3-8B, KV Cache Budget=10%2026.0648.31
PyramidKVBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.0648.02
H2OBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.0647.18
PyramidKVBackbone=Llama-3-8B, KV Cache Budget=10%2026.0647.13
SnapKVBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.0647.11
SnapKVBackbone=Llama-3-8B, KV Cache Budget=10%2026.0646.23
H2OBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.0644.52
H2OBackbone=Llama-3-8B, KV Cache Budget=10%2026.0643.84
Full KVBackbone=Llama-3.2-3B, KV Cache Budget=Full2026.0643.21
SEKVBackbone=Mistral-7B, KV Cache Budget=10%2026.0642.63
StreamingLLMBackbone=Qwen2.5-14B, KV Cache Budget=10%2026.0642.31
Full KVBackbone=Mistral-7B, KV Cache Budget=Full2026.0641.83
SentenceKVBackbone=Mistral-7B, KV Cache Budget=10%2026.0641.44
SemantiCacheBackbone=Mistral-7B, KV Cache Budget=10%2026.0641.17
ChunkKVBackbone=Mistral-7B, KV Cache Budget=10%2026.0640.83
PyramidKVBackbone=Mistral-7B, KV Cache Budget=10%2026.0639.74
SEKVBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.0639.47
StreamingLLMBackbone=Llama-3.1-8B, KV Cache Budget=10%2026.0639.17
SnapKVBackbone=Mistral-7B, KV Cache Budget=10%2026.0638.91
StreamingLLMBackbone=Llama-3-8B, KV Cache Budget=10%2026.0638.41
SentenceKVBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.0637.83
H2OBackbone=Mistral-7B, KV Cache Budget=10%2026.0637.63
SemantiCacheBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.0637.14
ChunkKVBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.0636.82
PyramidKVBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.0635.44
BLASSTModel=Qwen3-8B, Sparsity=~50%, Deployment Phase=Prefill Phase2025.1235.1
StreamingLLMBackbone=Mistral-7B, KV Cache Budget=10%2026.0634.82
BLASSTModel=Qwen3-8B, Sparsity=~75%, Deployment Phase=Prefill Phase2025.1234.4
SnapKVBackbone=Llama-3.2-3B, KV Cache Budget=10%2026.0634.18
Dense AttentionModel=Qwen3-8B, Sparsity=Dense, Deployment Phase=Prefill Phase2025.1233.6
Dense AttentionModel=Qwen3-8B, Sparsity=Dense, Deployment Phase=Prefill + Decode Phase2025.1233.6
BLASSTModel=Qwen3-8B, Sparsity=~50%, Deployment Phase=Prefill + Decode Phase2025.1233.3
BLASSTModel=Qwen3-8B, Sparsity=~75%, Deployment Phase=Prefill + Decode Phase2025.1233.1
BLASSTModel=Llama-3.1-8B, Sparsity=~50%, Deployment Phase=Prefill + Decode Phase2025.1232.4