Model	techniques
GLM-5.3-Flash used	Multi-Token Prediction used—
DeepSeek-V4.1-Flash core	DSpark coreThroughput-aware dynamic verification-length scheduling coreSpeculative decoding used—
Hy4-preview used	Multi-Token Prediction usedNEXTN speculative decoding usedSpeculative decoding used—
DeepSeek-V4-Flash-0731 used	DSpark usedChat Prefix Completion optionalSpeculative decoding mentioned—
NVIDIA-Nemotron-3-Ultra-550B-A55B core	Recursive shared MTP-head drafting coreSpeculative decoding defaultBest-of-N scaffolding usedEAGLE usedMulti-Token Prediction used—
MiMo-V2.6-Flash core	DFlash coreDistribution-matched draft-model fine-tuning usedThroughput-based draft-block sizing usedEAGLE optionalSpeculative decoding optional—
MiMo-V2.5 used	Multi-Token Prediction usedSpeculative decoding usedEAGLE optional—
GLM-5.3 used	Multi-Token Prediction used—
Hy3 core	Speculative decoding coreEAGLE optionalSpeculative sampling optional—
GLM-5.2 used	Multi-Token Prediction usedSpeculative decoding used—
DeepSeek-V3.2 used	Longest-trace selection usedMulti-stage candidate filtering usedTop-p and top-k sampling used—
DeepSeek-V4-Flash-Vision-Exp default	Same-checkpoint target and draft weights defaultChat Prefix Completion optionalDSpark optional—
DeepSeek-V4-Pro-0813 optional	Chat Prefix Completion optionalDSpark optional—
Gemma 4 31B core	Multi-Token Prediction coreSpeculative decoding coreStandardized sampling configuration defaultKV-cache sharing between drafter and target usedTop-k over token clusters used—
Inkling core	Multi-Token Prediction core—
Kimi K3 core	Fused recurrent replay kernel coreEAGLE-3-style draft-model fine-tuning used—
Laguna-S-2.1 optional	DFlash optional—
MiMo-V2.5-Pro used	Multi-Token Prediction usedSpeculative decoding usedEAGLE optionalMulti-layer EAGLE optional—
MiMo-V2.6-Pro core	DFlash coreDistribution-matched draft-model fine-tuning usedThroughput-based draft-block sizing used—
NVIDIA-Nemotron-3.5-Lightning-30B-A3B core	Speculative decoding coreBest-of-N scaffolding usedMulti-Token Prediction usedDFlash optionalDSpark optionalConcurrency-aware draft-length tuning evaluated—
Qwen3.5-397B-A17B used	MTP-1 speculative decoding usedPresence Penalty usedMulti-Token Prediction optionalNEXTN speculative decoding optionalTask-specific sampling parameters optional—
Qwen3.6-35B-A3B optional	Multi-Token Prediction optionalPresence Penalty optionalTask-specific sampling parameters optional—
Qwen3.8-Flash-Next used	NEXTN speculative decoding usedMulti-Token Prediction optional—
Step-3.7-Flash optional	EAGLE optionalMulti-layer EAGLE optionalSpeculative decoding optional—
gpt-oss-120b used	Grammar-constrained decoding usedBest-of-N scaffolding evaluated—

Model

techniques

GLM-5.3-Flash used

Multi-Token Prediction used—

DeepSeek-V4.1-Flash core

DSpark coreThroughput-aware dynamic verification-length scheduling coreSpeculative decoding used—

Hy4-preview used

Multi-Token Prediction usedNEXTN speculative decoding usedSpeculative decoding used—

DeepSeek-V4-Flash-0731 used

DSpark usedChat Prefix Completion optionalSpeculative decoding mentioned—

NVIDIA-Nemotron-3-Ultra-550B-A55B core