Intelligence
22
score
DeepSeek V3.2 Speciale scores 22 (estimated) on the Artificial Analysis Intelligence Index, placing …
vs class avg: 25
DeepSeek•Open weights•December 1, 2025
DeepSeek V3.2 Speciale
Intelligence, Performance & Price Analysis
Source:Artificial Analysis scrapeType:Open weights
Input Price#1/93
$0.00
Output Price#1/93
$0.00
Speed
not provided
This profile does not provide this metric.
Latency
not provided
This profile does not provide this metric.
Decision Overview
Ethen's intelligence-driven routing assessment for this model
Overall verdict
Strong general-purpose model
Suitable for most production tasks, but high-volume or repetitive work should still be compared against cheaper routes.
Best for
- Code generation
- Content analysis
- Structured extraction
- Customer-facing chat
Weak or unsuitable
- Extremely long-context tasks
- Ultra-low-latency requirements
Routing recommendation
1
Use the fit matrix to compare against cheaper routes when volume rises
Cost pressure
Competitive pricing — $0.00 / M input, $0.00 / M output. Cost pressure is low enough for sustained production use.
Model-fit matrix
Use this model where capability pays for itself; route away where volume, latency, or budget dominate.
| Use case | Fit | Reason | Routing note |
|---|---|---|---|
| Complex reasoning & agentic workflows | Strong | Intelligence score 22 supports capable reasoning, but very hard tasks may benefit from higher-tier models. | Good for most complex tasks; consider a frontier model for the hardest 10%. |
| Cost-sensitive pipelines | Excellent | Output pricing at $0.00 is very competitive for high-volume workloads. | Excellent for budget-constrained pipelines; enable caching to reduce costs further. |
Cost pressure
LowPricing is competitive — input $0.00, output $0.00. Suitable for sustained production use.
Cheaper substitutes
No cheaper substitute is supported by the current price data.
Route away when
- Extreme scale where even low costs matter
Highlights
Key benchmark results at a glance.
Intelligence
Intelligence position
Artificial Analysis Intelligence Index · Higher is better · Evaluation results measured independently by Artificial Analysis
DeepSeek V3.2 Speciale scores 22 (estimated) on the Artificial Analysis Intelligence Index, placing it below average among other open weight models of similar size (median: 25).
Speed
Speed position
Output tokens per second · Higher is better · Evaluation results measured independently by Artificial Analysis
Higher throughput supports faster interactive use.
Context, Cost & Pricing
Price/value position
Weighted average cost (USD) per Intelligence Index task · Lower is better · Evaluation results measured independently by Artificial Analysis
Pricing is competitive — input $0.00, output $0.00. Suitable for sustained production use.
Intelligence & Quality
Benchmark scores and quality indices measuring model capability.
AA-Omniscience Index
AA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct. · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Bars intentionally show many models at once so the page feels like a benchmark report rather than a dashboard.
Artificial Analysis Intelligence Index by Open Weights / Proprietary
Artificial Analysis Intelligence Index v4.1 incorporates 9 evaluations: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Artificial Analysis Intelligence Index
Artificial Analysis Intelligence Index v4.1 incorporates 9 evaluations: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Artificial Analysis Openness Index: Score
Openness Index assesses model openness on a 0 to 100 normalized scale (higher is more open) · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Bars intentionally show many models at once so the page feels like a benchmark report rather than a dashboard.
Intelligence
Artificial Analysis Intelligence Index · Higher is better · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Speed & Throughput
Output tokens per second and generation throughput.
Speed
Output tokens per second · Higher is better · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Speed should be read together with latency and cost; fast output does not always mean better user-perceived performance.
Context, Cost & Pricing
Context window size, cost per task, and token pricing.
Context Window
Context window: tokens limit · Higher is better · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Larger context windows enable complex multi-turn tasks but may increase latency and cost for long inputs.
Cost per Intelligence Index Task
Weighted average cost (USD) per Artificial Analysis Intelligence Index task, segmented by token type. Lower is better · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Ethen Gateway should route away from this model when the task is simple, repetitive, or likely to generate many output tokens.
Cost per Task
Weighted average cost (USD) per Intelligence Index task · Lower is better · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Ethen Gateway should route away from this model when the task is simple, repetitive, or likely to generate many output tokens.
Cost to Run Artificial Analysis Intelligence Index
Cost (USD) to run all evaluations in the Artificial Analysis Intelligence Index · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Higher values indicate stronger overall capability. Bars show how models compare on composite intelligence metrics.
Ethen Gateway should route away from this model when the task is simple, repetitive, or likely to generate many output tokens.
Pricing: Cache Hit, Input, and Output
Price (USD per M Tokens) · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Bars intentionally show many models at once so the page feels like a benchmark report rather than a dashboard.
Architecture & Scale
Model size, parameters, and architectural details.
Model Size: Total and Active Parameters
Comparison between total model parameters and parameters active during inference · Evaluation results measured independently by Artificial Analysis
Source and methodology are summarized below
Chart source and provenance are listed in Methodology & sources below.
Larger models generally perform better but cost more to serve. Active parameter count matters for inference efficiency.
Specifications
Technical Specifications
DeepSeek V3.2 Speciale scores 22 (estimated) on the Artificial Analysis Intelligence Index, placing it below average among other open weight models of similar size (median: 25).
DeepSeek V3.2 Speciale
- Model type
- Open weights
- Reasoning
- Yes
- Input modalities
- DeepSeek V3.2 Speciale supports text input.
- Output modalities
- DeepSeek V3.2 Speciale supports text output.
- Context window
- 130k tokens
- Open weights / source
- Yes, DeepSeek V3.2 Speciale is open weights. The model weights are publicly available and can be downloaded for self-hosting.
- Parameters
- DeepSeek V3.2 Speciale has 685 billion parameters (37 billion active).
- Active parameters
- DeepSeek V3.2 Speciale is a Mixture of Experts (MoE) model with 685 billion total parameters, but only 37 billion active parameters are used during inference.
- License
- DeepSeek V3.2 Speciale is released under the MIT license. This license allows commercial use.
Methodology & Provenance
This page is rendered from the normalized profile and page JSON for DeepSeek V3.2 Speciale.
Benchmark values are preserved as normalized; only layout, disclosure ordering, and typography are adjusted for readability.
Source file: md/scrape/artificial_analysis_model_pages/artificialanalysis_ai__models__deepseek-v3-2-speciale__9a9a50c6dc.html · Canonical URL: https://artificialanalysis.ai/models/deepseek-v3-2-speciale
Normalized at: 2026-07-08T03:24:56+00:00 Technical specs, comparison summary, and FAQs are passed through from the normalized data without rewriting.
Ranks shown on summary cards come directly from the normalized profile. No rank or class display is synthesized in the renderer.
Visible FAQ answers and FAQ JSON-LD are emitted from the same normalized FAQ array, so structured data stays aligned with the rendered accordion.
Speed: https://artificialanalysis.ai/models/deepseek-v3-2-speciale
Frequently Asked Questions
DeepSeek V3.2 Speciale was released on December 1, 2025.
DeepSeek V3.2 Speciale was created by DeepSeek.
DeepSeek V3.2 Speciale scores 22 (estimated) on the Artificial Analysis Intelligence Index, placing it below average among other open weight models of similar size (median: 25).