Skip to content

DeepSeek · Chat

DeepSeek V3.2 Exp

DeepSeek 🇨🇳ChatDeepSeek familyOpen weights. One of its sources says the weights are publicly available.
From Meganova's listing:
Experimental version with sparse-attention architecture for long-context efficiency
llminfo IQ
123.7
Maker's price
$0.305 blended / 1M
Max context
164K
Listings
23

Evidence and sources

🤗 deepseek-ai/DeepSeek-V3.2-Expllminfo IQ 123.7, ranked among chat models. From 179 benchmarks. Plenty of data behind it.How llminfo IQ is computedCoding IQ 97.5, 95% interval 90–105. From 9 benchmark families across 3 sources. A fair amount of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.How Coding IQ is computedGood value (68.3). Cheaper per task than its benchmark results predict. From 12 benchmarks with measured task costs. Plenty of data behind it.How Value for money worksReleased 29 Sep 2025 Training-data cutoff, as its sources state it Maker HQ: CN (according to DeepSeek's own website)
All rankings: #176 on NanoGPT recently #192 on OpenRouter today #202 on OpenRouter this week #213 on OpenRouter this monthSources: LMArena data, SiliconFlow data, Models.dev data, OpenRouter data docs, MegaNova data, Hugging FacePrimary data from LMArena from SiliconFlow: name from Models.dev: weights availability from OpenRouter: knowledge cutoff from MegaNova: description from Hugging Face: release date

Advantages and trade-offs

Compared with 10 chat models, picked mainly by popularity and IQ

Advantages

Open weights available

The weights are public. 3 of the 8 peers that say either way have open weights.

Trade-offs

We only make a claim when there's enough data on models of the same type. Updated 22 Nov, 04:34 UTC.Prices come from the model's maker where they sell it, otherwise their official partner, otherwise the market median. This comparison uses 10 maker's prices.

Input & output

as stated by the model's providers
Text
Text

Capabilities

Supported operations and provider-stated features

ReasoningFunction callingStructured output

Listings

 DeepSeekdeepseek-v3.2-expSource: GenAI Prices dataNot published by the provider$0.27$0.41
 Zenmuxdeepseek/deepseek-v3.2-expSource: Models.dev data163K$0.22$0.33
 Atlasclouddeepseek-ai/DeepSeek-V3.2-ExpSource: AtlasCloud data164K$0.26cache read $0.13$0.38
 Meganovadeepseek-ai/DeepSeek-V3.2-ExpSources: MegaNova data, Models.dev data164K$0.26cache read $0.13$0.38
 Flowdotdeepseek/deepseek-v3.2-expSource: FlowDot data164K$0.27$0.41
 Hugging FaceInference by Novita AIdeepseek-ai/DeepSeek-V3.2-ExpSource: GenAI Prices data164K$0.27$0.41
 Kilodeepseek/deepseek-v3.2-expSources: Kilo AI Gateway data, Models.dev data164K$0.27$0.41
 Nous Portaldeepseek/deepseek-v3.2-expSource: Nous Portal Inference Gateway data164K$0.27$0.41
 OpenRouterdeepseek/deepseek-v3.2-expSources: GenAI Prices data, Models.dev data, OpenRouter data docs164K$0.27$0.41
 Gmicloudgmi/deepseek-ai/DeepSeek-V3.2-ExpSource: LiteLLM data164K$0.27$0.41
Show all 20 listings

Showing 10 of 20: the maker's own listing first, then the cheapest per 1M input tokens from each provider.

Also sold by 302ai, Alibaba Cloud (China), NanoGPT, Novita AI, Qiniu, SiliconFlow.

3 listings without a price are hidden.

Benchmarks

Named effort levels run from lowest to highest. Select a result to see its source details.

Human preference

Arena ELO43 related benchmarks · 73 results
Source: LMArena data · Nov 2026
Arena ELO
Chinese
Creative Writing
English
Exclude Ties
Expert
Show 61 more resultsHide 61 results
French
German
Hard Prompts
Hard Prompts English
Industry Entertainment and Sports and Media
Industry Life and Physical and Social Science
Industry Mathematical
Industry Medicine and Healthcare
Industry Software and It Services
Industry Writing and Literature and Language
Japanese
Korean
Longer Query
Multi Turn
Non English
Polish
Russian
Design Arena EloSource: Design Arena Leaderboards data · Nov 2026

Multilingual

IndicParam42 related benchmarks · 42 results
Show 36 more resultsHide 36 results
LaoBench16 related benchmarks · 32 results
Show 20 more resultsHide 20 results
Lao-500 judge difference Δ(Gemini-2.5-Pro − Qwen3-Max)
Source: LaoBench (arXiv:2511.11334v3) data
By mode · table: 10 · statistic: delta
Lao-500 win rate vs GPT-5-High (Gemini-2.5-Pro judge)
Source: LaoBench (arXiv:2511.11334v3) data
By mode · table: 10 · judge: Gemini-2.5-Pro
Lao-500 win rate vs GPT-5-High (Qwen3-Max judge)
Source: LaoBench (arXiv:2511.11334v3) data
By mode · table: 10 · judge: Qwen3-Max

Long context

Coding

Reasoning

Agentic

General

Math

AIME 2025Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
HMMT 2025Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Factuality

SimpleQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Instruction following

Science

GPQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Knowledge

MMLU-ProSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Other

Data credits: Data by LLM Stats · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · LMArena, CC BY 4.0 (licence) · Maheshwari et al., IndicParam, arXiv:2512.00333v3. Cited numeric scores under the owner's 2026-10-10 task-only restriction scope decision; no results licence established. Tasks, prompts and generations excluded. · LaoBench, BAAI et al., arXiv:2511.11334v3. Cited numeric results only. arXiv non-exclusive distribution notice establishes no results reuse grant; dataset Apache licence is not a results grant. · EQ-Bench (Samuel J. Paech) — cite paech2023eqbench · Vectara Hallucination Leaderboard, Apache-2.0 (licence)

Also known as: deepseek-ai/deepseek-v3.2-exp, deepseek-ai/deepseek-v3.2-exp-thinking, deepseek-v3-2-exp, deepseek-v3.2-exp-nothinking, deepseek-v3.2-exp-thinking, deepseek/deepseek-v3.2-exp, deepseek/deepseek-v3.2-exp-thinking, gmi/deepseek-ai/deepseek-v3.2-exp, novita/deepseek/deepseek-v3.2-exp

Like this one. Tap or hover any dotted value, label or icon to see where it comes from and how it is measured. open details when tapped or hovered. How this data works