DeepSeek V3.2

Honghao Wang

— 2 min read
DeepSeek V3.2

DeepSeek-V3.2 Release Overview

DeepSeek-V3.2 (via Hacker News) — DeepSeek has released two new open-weight (MIT licensed) models:

Key facts:

---

Model Differences

From the official technical report:

> DeepSeek-V3.2 integrates reasoning, agent, and human alignment data distilled from specialists, undergoing thousands of reinforcement learning steps to reach final checkpoints.

> DeepSeek-V3.2-Speciale is an experimental variant trained exclusively on reasoning data with reduced length penalty during RL. It adopts datasets and reward methods from DeepSeekMath-V2 (Shao et al., 2025) to boost mathematical proof capabilities.

Achievements:

  • Both models achieved gold in the 2025 International Mathematical Olympiad benchmark — a notable metric for AI training teams.

---

Relation to DeepSeek-Math-V2

Last week’s coverage: DeepSeek-Math-V2

  • Similar mathematical performance enhancements
  • Especially strong in reasoning and extended problem solving
  • Speciale variant demonstrates greater focus on long-form, precise proofs.

---

Test Prompt: Creative SVG Generation

Test platform: OpenRouter chat interface

Prompt: _"Generate an SVG of a pelican riding a bicycle."_

DeepSeek-V3.2 Output

  • Response: short reasoning chain
  • Prompt setup:
Let's assume the following:
Wheel radius: 40
Distance between wheel centers: 180
Seat height: 60 (above the rear wheel center)
Handlebars: above the front wheel, extending backward and upward.
We'll set the origin at the center of the rear wheel.
We'll create the SVG with a viewBox that fits the entire drawing.
Let's start by setting up the SVG.

Illustration:

image

---

DeepSeek-V3.2-Speciale Output

Illustration:

image

---

Open-Weight AI for Creative Workflows

Models like DeepSeek-V3.2 exemplify the rapid expansion of open-weight AI, enabling:

  • Experimentation without restrictive licenses
  • Use in both research and creative applications
  • Integration into multi-platform publishing workflows

---

AiToEarn — AI Content Generation & Monetization

AiToEarn官网 is an integrated ecosystem for creators:

Features:

  • Generate AI content — visual or text — using open-source models
  • Publish simultaneously to:
  • Douyin, Kwai, WeChat, Bilibili, Rednote
  • Facebook, Instagram, LinkedIn, Threads
  • YouTube, Pinterest, X (Twitter)
  • Track analytics across channels
  • Model rankings available via AI模型排名
  • Monetization tools to streamline earnings from creative output

Use case:

Whether generating mechanical schematics or whimsical illustrations like the pelican bicycle, AiToEarn supports global distribution and engagement tracking in an open-source monetization framework.

---

Explore more:

Read more

Translate the following blog post title into English, concise and natural. Return plain text only without quotes. 哈佛大学 R 编程课程介绍

Harvard CS50: Introduction to Programming with R Harvard University offers exceptional beginner-friendly computer science courses. We’re excited to announce the release of Harvard CS50’s Introduction to Programming in R, a powerful language widely used for statistical computing, data science, and graphics. This course was developed by Carter Zenke.