DeepSeek-V4-Pro Officially Released

We are thrilled to announce that DeepSeek V4 Pro and V4 Flash are now fully open source — a milestone that ushers open-source large models into a new era.

Key Highlights

• 1M token context: Process ultra-long documents and codebases without chunking or truncation
• Agent capabilities: 73.6% on the MCPAtlas benchmark, with robust multi-tool orchestration
• Vision-primitive mode: Precise spatial reasoning with up to 60% lower token consumption
• DSpark acceleration: Speculative decoding framework delivering up to 85% faster responses

Model Specifications

Model	Parameters	Context	Best For
V4 Pro	671B MoE	1M	Complex reasoning, research, enterprise workloads
V4 Flash	37B	128K	Fast responses, everyday chat, edge deployment

Real-World Performance

We stress-tested V4 Pro in production-like scenarios:

• Long-document processing: A 100-page (~80K token) technical PDF was fully summarized in 12 seconds with zero missed key conclusions.
• Coding ability: Codeforces rating of 3206 — outperforming roughly 95% of human competitive programmers.
• Reasoning accuracy: 90.1% on GPQA Diamond graduate-level reasoning, ahead of comparable closed-source models.

Open-Source License

Both V4 Pro and V4 Flash are released under the MIT license, allowing commercial use. Visit DeepSeek V4 Pro for API docs and tutorials.

Get Started with V4 Pro in 5 Minutes

Basic API Call

from openai import OpenAI

client = OpenAI(
    api_key="your-api-key",
    base_url="https://api.deepseek.com/v1"
)

response = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[{"role": "user", "content": "Write a quicksort in Python"}],
    temperature=0.7
)
print(response.choices[0].message.content)

10 Practical Prompt Examples

Code Generation

Build a concurrent API gateway in Go with rate limiting and circuit breaking

Paper Analysis

Summarize this paper's core contributions, methodology, and limitations: [paste abstract]

Creative Writing

Write a 500-word sci-fi short story about AI awakening, third-person perspective

Long-Document Summary

This is an 80-page technical report. Extract: 1) core conclusions 2) key data 3) risks to validate

Bug Diagnosis

This Python code throws KeyError under concurrency. Find root cause and fix: [paste code]

Data Analysis

Given this CSV, compute Q2 QoQ growth by region and suggest visualizations: [paste data]

Performance Comparison

Model	Code Quality	Speed	Context
V4 Pro	⭐⭐⭐⭐⭐	45 tok/s	1M
V4 Flash	⭐⭐⭐⭐	112 tok/s	128K
GPT-4o	⭐⭐⭐⭐	38 tok/s	128K

FAQ

Q: Where do I get an API key?

A: Register at platform.deepseek.com for free credits.

Q: V4 Pro or V4 Flash?

A: Choose Pro for complex reasoning, long documents, and Agent tasks; Flash for high-concurrency chat and cost-sensitive workloads.

Q: Can I use it commercially?

A: Yes. The MIT license permits free commercial use, and the API offers trial credits.

---

📅 Updated July 2, 2026. Data source: DeepSeek 2026 official technical report.