# DeepSeek-V3.2-Speciale

DeepSeek-V3.2-Speciale is an AI model released by DeepSeek on Dec 1 2025. It has 671B parameters, a 128k context window and open weight. Tracked results include 85.7% on GPQA Diamond, 88.7% on LiveCodeBench and 30.6% on Humanity's Last Exam (text only, no tools).

## Facts

| Field | Value |
| --- | --- |
| Model | DeepSeek-V3.2-Speciale |
| Developer | DeepSeek |
| Release date | Monday, Dec 1 2025 |
| Licensing | Open Weight |
| Parameters | 671B |
| Context window | 128k |

## Tracked benchmark scores

| Benchmark | Score | Source | What it measures |
| --- | --- | --- | --- |
| LiveCodeBench | 88.7% | Lab | Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better. |
| Humanity's Last Exam (text only, no tools) | 30.6% | Lab | accuracy on the 2,158 text-only questions without tools |
| AIME 2025 | 96% | Lab | single-answer accuracy on the 2025 invitational mathematics exam without tools |
| GPQA Diamond | 85.7% | Lab | Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. |

## About DeepSeek-V3.2-Speciale

DeepSeek-V3.2-Speciale launched alongside V3.2 on December 1, 2025 as a variant that spent more computation on reasoning. DeepSeek relaxed output-length constraints to improve performance on demanding mathematics and programming tasks.

DeepSeek published the weights and initially offered a temporary API endpoint without tool calls. Its benchmark values belong to Speciale's own column in the technical report, including the text-only, no-tools version of Humanity's Last Exam.

## Questions and answers

### When was DeepSeek-V3.2-Speciale released?

DeepSeek-V3.2-Speciale was released by DeepSeek on Monday, Dec 1 2025.

### Who made DeepSeek-V3.2-Speciale?

DeepSeek-V3.2-Speciale was built by DeepSeek. Chinese AI lab known for efficient, open-weight models. Gained attention for strong performance at lower cost.

### What benchmark scores did DeepSeek-V3.2-Speciale get?

DeepSeek-V3.2-Speciale reports 4 tracked benchmark scores — LiveCodeBench: 88.7%; Humanity's Last Exam (text only, no tools): 30.6%; AIME 2025: 96%; GPQA Diamond: 85.7%. Tracked scores may come from lab reports or independent benchmarks; source details accompany the benchmark data.

### What is the context window of DeepSeek-V3.2-Speciale?

DeepSeek-V3.2-Speciale has a context window of 128k. Providers may specify separate input and output token limits.

### How many parameters does DeepSeek-V3.2-Speciale have?

DeepSeek-V3.2-Speciale is reported at 671B parameters.

### Is DeepSeek-V3.2-Speciale open source?

Partly. DeepSeek-V3.2-Speciale is an open-weight model: the trained weights are free to download and run locally or on your own infrastructure, but the training data and code are not fully released and the license may restrict some commercial uses. It is not open source in the strict sense.

### What came before and after DeepSeek-V3.2-Speciale?

DeepSeek's previous tracked release was DeepSeek-V3.2 on Dec 1 2025. It was followed by DeepSeek-V4-Pro on Apr 24 2026.


---

Canonical page: https://aireleasetracker.com/model/deepseek/deepseek-v3.2-speciale
Site index: https://aireleasetracker.com/llms.txt
Source: AI Release Tracker (https://aireleasetracker.com). Most benchmark scores come from lab launch material; gathered results identify the leaderboard that published them.
