List your company

Scientific Reasoning on GPQA Diamond (Acc, Tok, CR, Cost)

65.25AccuracyFull
14.0327.327540.62553.9225Apr 6, 2026Apr 15, 2026Apr 25, 2026May 4, 2026May 14, 2026May 23, 2026Jun 2, 2026

Evaluation results

MethodLinks
FullModel=Qwen3-14B, KV Cache Budget=100%2026.0665.25———
VanillaModel=Qwen3-14B2026.0465.28,4471008,447
RCPDModel=Qwen3-14B2026.0465.28,09895.98,433
SeerAttention-RModel=Qwen3-14B, KV Cache Budget=~25%2026.0661.68———
R-KVModel=Qwen3-14B, KV Cache Budget=~25%2026.0659.09———
EATModel=Qwen3-14B2026.0458.36,35575.26,533
VASE-ATTNVModel=Qwen3-14B, KV Cache Budget=~25%2026.0657.39———
VASE-DKVModel=Qwen3-14B, KV Cache Budget=~25%2026.0656.76———
FullModel=Qwen3-4B, KV Cache Budget=100%2026.0656.19———
DEERModel=Qwen3-14B2026.04563,76344.53,799
VanillaModel=Qwen3-4B2026.0455.89,5361009,536
RCPDModel=Qwen3-4B2026.0455.89,42498.89,825
Think or NotModel=Qwen3-4B2026.0454.16,8657210,371
CoDE-StopModel=Qwen3-14B2026.0453.83,128373,159
VanillaModel=Nemotron-8B2026.0453.27,3181007,318
RCPDModel=Nemotron-8B2026.0453.27,243997,558
DEERModel=Qwen3-4B2026.04536,41267.26,496
CoDE-StopModel=Qwen3-4B2026.0452.86,06463.66,144
Think or NotModel=Nemotron-8B2026.0452.86,8049311,142
DEERModel=Nemotron-8B2026.0451.75,88380.45,967
CoDE-StopModel=Nemotron-8B2026.0450.95,634775,715
SeerAttention-RModel=Qwen3-4B, KV Cache Budget=~25%2026.0649.94———
R-KVModel=Qwen3-4B, KV Cache Budget=~25%2026.0649.05———
EATModel=Qwen3-4B2026.04496,45967.76,670
EATModel=Nemotron-8B2026.04496,33486.66,535
CurDKVModel=Qwen3-14B, KV Cache Budget=~25%2026.0648.48———
VASE-ATTNVModel=Qwen3-4B, KV Cache Budget=~25%2026.0647.98———
Think or NotModel=DeepSeek-R12026.0447.59,05698.718,243
VanillaModel=DeepSeek-R12026.0447.49,1771009,177
RCPDModel=DeepSeek-R12026.0447.49,1781009,527
ExpThinkBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0547.22,870.9——
DEERModel=DeepSeek-R12026.04478,03287.58,468
CoDE-StopModel=DeepSeek-R12026.0446.97,89886.18,318
EATModel=DeepSeek-R12026.0445.77,94886.68,375
Answer ConvergenceModel=Qwen3-14B2026.0444.78309.81,060
LaserBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0544.13,492.9——
VASE-DKVModel=Qwen3-4B, KV Cache Budget=~25%2026.0643.94———
LC-R1Backbone=DeepSeek-R1-Distill-Qwen-7B2026.0540.84,295.6——
JETBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0540.36,491.1——
Answer ConvergenceModel=Qwen3-4B2026.0439.19479.91,210
AdaptThinkBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0538.35,225——
CurDKVModel=Qwen3-4B, KV Cache Budget=~25%2026.0637.94———
AutoThinkBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0537.95,431.5——
SnapKVModel=Qwen3-4B, KV Cache Budget=~25%2026.0636.87———
VanillaBackbone=DeepSeek-R1-Distill-Qwen-7B2026.0535.27,288.7——
Answer ConvergenceModel=Nemotron-8B2026.0433.91,23116.81,656
Answer ConvergenceModel=DeepSeek-R12026.0433.32,62928.64,394
ExpThinkBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0525.43,259.4——
AutoThinkBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.05255,324——
AdaptThinkBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0524.44,013.8——
JETBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0523.96,566.8——
LC-R1Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.05235,285.2——
LaserBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.0521.85,136.9——
VanillaBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.05168,514.3——