Adjustable thinking AI models: set the reasoning budget per request

Prices from official vendor price lists · October 2026
79 models that let you dial reasoning depth up or down — a direct control on both quality and cost. Output prices run from $0.13 to $50 per 1M tokens across the 73 models with a paid per-token price; 48 have a free tier and 41 ship open weights.
Adjustable thinking
Services found: 79
#AI modelCapabilitiesAPI $/1M (in / out)Context windowPrice / moOutput $/1MOverviewVisit site
21

Gemini 3.6 Flash

Google
Text & chatVisionAudio
$0.75 / $3.751.05M tokensAPI only3.75View
22

Qwen3.8 27B

Qwen
Text & chatVisionVideo
$0.42 / $31M tokensAPI only3View
23

DeepSeek V4 Pro 0423

DeepSeek
Text & chatReasoningAdjustable thinking
$0.95526 / $1.910521.05M tokensAPI only1.911View
24

Claude Sonnet 4.6

Anthropic
Text & chatVisionReasoning
$3 / $151M tokensAPI only15View
25

Qwen3.7 Max

Qwen
Text & chatReasoningAdjustable thinking
$2.5 / $7.51M tokensAPI only7.5View
26

Qwen3.7 Plus

Qwen
Text & chatVisionReasoning
$0.32 / $1.281M tokensAPI only1.28View
27

Grok 4.3

xAI
Text & chatVisionReasoning
$1.25 / $2.51M tokensAPI only2.5View
28

DeepSeek V4 Flash 0423

DeepSeek
Text & chatVisionReasoning
$0.088606 / $0.1772121.05M tokensAPI only0.177View
29

Gemini 3.5 Flash-Lite

Google
Text & chatVisionAudio
$0.3 / $2.51.05M tokensAPI only2.5View
30

Qwen3.6 27B

Qwen
Text & chatVisionVideo
$0.3 / $2262.1K tokensAPI only2View
31

Claude Sonnet 4.5

Anthropic
Text & chatVisionReasoning
$3 / $151M tokensAPI only15View
32

Qwen3.5 397B A17B

Qwen
Text & chatVisionVideo
$0.55 / $3.5262.1K tokensAPI only3.5View
33

Claude Haiku 4.5

Anthropic
Text & chatVisionReasoning
$1 / $5200K tokensAPI only5View
34

Qwen3.5-122B-A10B

Qwen
Text & chatVisionVideo
$0.29 / $2.4262.1K tokensAPI only2.4View
35

Qwen3 Coder Next

Qwen
Text & chatCodingReasoning
$0.12 / $0.8262.1K tokensAPI only0.8View
36

Gemini 3.1 Flash-Lite

Google
Text & chatVisionAudio
$0.25 / $1.51.05M tokensAPI only1.5View
37

Gemini 3.5 Transcribe

Google
AudioAdjustable thinkingSpeech-to-text
API onlyView
38

Gemini 3.5 Transcribe Live

Google
AudioAdjustable thinkingSpeech-to-text
API onlyView
39

Grok 4.20 Reasoning

xAI
Text & chatVisionReasoning
$1.25 / $2.51M tokensAPI only2.5View
40

Grok 4.20 Non-Reasoning

xAI
Text & chatVisionReasoning
$1.25 / $2.51M tokensAPI only2.5View

Browse by need

Reasoning models169 modelsCoding assistants17 modelsVision and images156 modelsLong context 1M+82 modelsFree tier available80 modelsOpen weights88 models

Eighty-five models let you set how long the model deliberates before answering. Because thinking tokens bill as output, that dial is a direct cost control: the same model can be cheap and quick or slow and thorough depending on a single parameter.

The Claude family exposes it most clearly, and Grok, Qwen3 and DeepSeek V4 offer their own variants - DeepSeek switches between thinking and non-thinking modes outright. Set the dial per task rather than per application: minimal for classification and extraction, generous for architecture decisions and debugging.

Quality as a dialThe same model can be cheap and quick or slow and thorough on demand.
Per task, not per appExtraction wants minimal depth; architecture decisions want maximum.
Direct cost controlBecause thinking bills as output, the dial moves the invoice immediately.

Models with this capability

The 79 models below all support this capability. Prices are input and output per 1M tokens from the vendor official price list; for dirhams multiply by 3.6725. Tap a card to open the model page.

Claude Fable 5.1
#1
Anthropic
in $10out $501M tokens
GPT-6.1 Sol
#2
OpenAI
in $2out $101.05M tokens
GPT-6 Sol
#3
OpenAI
in $2out $101.05M tokens
GPT-6 Luna
#4
OpenAI
in $0.1out $0.51.05M tokens
Claude Opus 5.5
#5
Anthropic
in $4out $201M tokens
Claude Sonnet 5.5
#6
Anthropic
in $2out $101M tokens
Grok 4.7
#7
xAI
in $2out $6500K tokens
DeepSeek V4.1 Flash
#8
DeepSeek
in $0.3out $1.21M tokens

How to choose

  • Profile your task mix before fixing a default depth.
  • Use minimal settings for anything deterministic — parsing, routing, classification.
  • Reserve deep settings for ambiguous problems where a wrong answer is expensive.
  • Log which depth produced which outcome so the setting can be tuned with evidence.

Common mistakes

One global settingA single depth for every request overpays on easy work and underdelivers on hard work.
Turning thinking off entirelyOn genuinely hard tasks it costs more in retries than it saves in tokens.

Who this suits

Cost-sensitive productsFixed budgets with variable task difficulty.
Agent buildersCapping deliberation keeps multi-step runs from spiralling.
Mixed workloadsOne model serving both trivial and demanding requests.

Related filters

Reasoning169 models that think before answering
Frontier tier41 flagship models, the best each vendor sells
Tool calling166 models that call your functions
Prompt caching147 models with cheap repeated context
All 239 modelsThe full rating with filters and prices

Frequently asked questions

How does the dial save money?

Thinking tokens bill as output. A shallow setting on the same model can cost a fraction of a deep one.

Which models expose it?

The Claude family most clearly, plus Grok, Qwen3 and DeepSeek V4, which switches thinking off entirely.

Should it be set per app or per task?

Per task. Minimal for extraction and classification, generous for architecture decisions and debugging.

Which vendors expose the control?

Anthropic, xAI, Alibaba and DeepSeek, each with its own naming.

Is low depth the same as a smaller model?

No. A large model thinking briefly still brings its full knowledge to the answer.

Frequently searched

By capabilityReasoningCodingVisionLong context 1M+Free tierOpen weightsLow latencyRealtime voiceImage generation
By vendorAnthropic (13)OpenAI (30)Google DeepMind (17)xAI (8)Meta (6)DeepSeek (9)Mistral AI (11)Alibaba (Qwen) (34)Z.ai (12)Moonshot AI (5)
Popular modelsClaude Fable 5.1GPT-6 AstraClaude Opus 5Claude Fable 5GPT-5.6 SolGLM 5.3Grok 4.6Kimi K3GPT-5.6 Terra
By deliveryBatch APIPrompt cachingPriority tierFrontier tierProprietary APITool callingStructured JSONWeb searchMultilingual

Ratings 0

Avg. score 0

Rate this page

Help us improve