AI Research #DeepSeek #AI model #benchmarks #China

DeepSeek V4 Outperforms Claude Opus 4.6 in Benchmarks

Chinese AI company DeepSeek launches V4 model with superior efficiency, available under MIT License.

Sunday April 26, 2026

TL;DR — Quick Answer

Chinese AI company DeepSeek has launched its V4 model, outperforming Claude Opus 4.6 in various benchmarks while offering superior efficiency. The model ships in two MoE variants, V4-Pro with 1.6 trillion parameters and V4-Flash with 284 billion, both trained on 32 trillion tokens and available under the MIT License.

Key Takeaways

→ DeepSeek-V4-Pro has 1.6 trillion parameters with 49 billion active
→ DeepSeek-V4-Flash has 284 billion parameters with 13 billion active
→ Both variants are MoE models trained on 32 trillion tokens
→ 1 million token context length with V4-Pro-Max topping HLE and Codeforces
→ Processes tasks in 1/3.7th the time of V3.2 Pro and 1/9.8th for Flash
→ Both versions are available on Hugging Face under the MIT License

DeepSeek V4 Outperforms Claude Opus 4.6 in Benchmarks — AI news article illustration

DeepSeek has launched its V4 AI model, outperforming Claude Opus 4.6 in various benchmarks while offering superior efficiency.

Model Variants

DeepSeek-V4-Pro — 1.6 trillion parameters, 49 billion active
DeepSeek-V4-Flash — 284 billion parameters, 13 billion active
Both are MoE (Mixture of Experts) models trained on 32 trillion tokens

Performance

1 million token context length
V4-Pro-Max tops benchmarks in HLE and Codeforces categories
Processes tasks in 1/3.7th the time of V3.2 (Pro) and 1/9.8th (Flash)

Availability

Both versions are available on Hugging Face under the MIT License, enabling broad adoption.

Market Impact

The release intensifies competition with OpenAI and Anthropic, especially given DeepSeek’s aggressive pricing.

Written by Massin BSN

Frequently Asked Questions

What is DeepSeek V4?

A new AI model family from DeepSeek that outperforms Claude Opus 4.6 on various benchmarks with better efficiency.

What variants does DeepSeek V4 come in?

V4-Pro with 1.6 trillion parameters and V4-Flash with 284 billion, both mixture-of-experts models.

Is DeepSeek V4 open source?

Both versions are available on Hugging Face under the MIT License.

How fast is DeepSeek V4?

It processes tasks in a third of the time of V3.2 Pro and about a tenth of the time for Flash.

AI Research