taxonomy node · level 2

reasoning control

42 methods filed at this node or below it, from the sources of 25 models.

inference & serving :: reasoning control

Matching aids for the classifier: reasoning effort control; thinking mode; test-time scaling; turn budget capping; adaptive reasoning; selectable reasoning depths.

By model

Which of this branch's techniques each model's own documents describe, and how strongly. Under each model: its strongest treatment anywhere in the branch.

Model	techniques
GLM-5.3-Flash core	Always-on thinking mode coreConfigurable reasoning effort defaultReasoning-effort resolution and system-prompt injection defaultclear_thinking chat-template parameter optional—
DeepSeek-V4.1-Flash core	Capped linear reasoning-token length deduction coreConfigurable reasoning effort coreEffort-dependent exponential token-penalty schedule coreDeployment-time scalar effort control defaultReward-optimized preferred reasoning length evaluatedTest-time compute scaling evaluated—
Hy4-preview default	Chain-of-thought reasoning defaultConfigurable reasoning effort used—
DeepSeek-V4-Flash-0731 used	Configurable reasoning effort used—
NVIDIA-Nemotron-3-Ultra-550B-A55B used	Chain-of-thought reasoning usedConfigurable reasoning effort usedGenerate-verify-refine loop usedMedium-effort reasoning mode usedTurn budget capping usedTurn-aware prompting usedInference-time reasoning budget control optional—
DeepSeek-V4-Flash used	Cross-turn persistent reasoning history usedThink-tag response formatting usedThinking mode selection optional—
GLM-5.3 default	clear_thinking chat-template parameter defaultConfigurable reasoning effort default—
Hy3 default	Configurable reasoning effort defaultReasoning parser usedChain-of-thought reasoning optional—
GLM-5.2 default	Configurable reasoning effort defaultDefault thinking mode usedMaximum thinking effort optional—
MiniMax-M3 used	Test-time compute scaling usedAdaptive reasoning optionalConfigurable reasoning effort optionalThinking mode selection optional—
DeepSeek-V3.2 used	Generate-verify-refine loop usedreasoning-enabled inference toggle optionalDefault thinking mode evaluatedParallel test-time compute scaling evaluatedParallel-fewest-step sampling evaluatedSerial test-time compute scaling through context management evaluated—
DeepSeek-V4-Flash-Vision-Exp optional	Configurable reasoning effort optional—
DeepSeek-V4-Pro used	Configurable reasoning effort usedCross-turn persistent reasoning history usedThink-tag response formatting usedThinking mode selection optional—
DeepSeek-V4-Pro-0813 optional	Configurable reasoning effort optional—
Gemma 4 31B core	Control-token-enabled thinking mode coreDefault thinking mode coreConfigurable thinking or reasoning mode optionalStep-by-step thinking mode optional—
Inkling used	Configurable reasoning effort usedControllable thinking effort via system message and per-token cost used—
Kimi K3 core	Always-on thinking mode coreChain-of-thought reasoning defaultConfigurable reasoning effort defaultMaximum thinking effort defaultNatural-language reasoning-effort option message usedPer-problem reasoning-budget control usedstage-wise curriculum over reasoning-effort budget multiplier usedThinking mode selection via generation prefix used—
Laguna-S-2.1 core	Interleaved thinking between tool calls coreConfigurable reasoning effort defaultMaximum thinking effort defaultThinking mode with automatic test-time compute budget defaultCross-turn persistent reasoning history usedThinking modes (off and max) usedUser-configurable thinking-effort control not used—
MiMo-V2.6-Pro used	Chain-of-thought reasoning used—
NVIDIA-Nemotron-3.5-Lightning-30B-A3B used	Inference-time reasoning budget control used—
Qwen3.5-397B-A17B default	Default thinking mode defaultDisabling reasoning via chat-template configuration optionalTask- and mode-specific sampling parameter recommendations optionalTask-appropriate maximum output length optionalQwen3 soft thinking switch not used—
Qwen3.6-35B-A3B default	Default thinking mode defaultInterleaved thinking between tool calls defaultThinking mode selection defaultCross-turn persistent reasoning history usedDisabling reasoning via chat-template configuration optional—
Qwen3.8-Flash-Next default	Default thinking mode default—
Step-3.7-Flash used	Reasoning parser usedConfigurable reasoning effort optional—
gpt-oss-120b core	Chain-of-thought reasoning coreConfigurable reasoning effort used—

reasoning control

In this branch 42

filed here 42

By model