Arena.ai Ranks DeepSeek V4 Pro Alongside Proprietary Frontier Models for Agentic Coding

ArenaArena

· Updated

Arena.ai confirmed that DeepSeek V4 Pro has entered its leaderboards as the #3 open model for coding and #2 for text. The rankings verify that this open-weight model matches the performance of closed-source systems like GPT-5.4-high in specialized agentic web development and healthcare tasks.

Arena.ai, a community-driven platform for evaluating AI models through human preference, released the first official rankings for the DeepSeek V4 family. The flagship DeepSeek V4 Pro is a Mixture-of-Experts (MoE) model (activating only a fraction of parameters per request) featuring 1.6 trillion total parameters.
Total parameters (Pro)
1.6T parameters
Activated parameters (Pro)
49B parameters
Total parameters (Flash)
284B parameters
Activated parameters (Flash)
13B parameters
Context window
1M tokens
Code Arena rank (Pro)
#3 open model
Text Arena rank (Pro)
#2 open model

The results mirror DeepSeek's V4 launch by placing the Pro variant at #14 overall in both the Code and Text Arenas. In agentic web development, the model showed parity with proprietary systems, matching OpenAI's GPT-5.5 performance. This confirms that open-weight models can compete at the frontier.

You can now deploy these models for complex reasoning workflows, such as medical analysis where V4 Pro currently ranks #1. While the Pro model handles high-end tasks, the smaller DeepSeek V4 Flash offers a cost-effective alternative, following a pattern seen in Xiaomi's MiMo-V2.5-Pro. Both are available for local hosting.

View the full update on arena.ai
Arena.ai
Arena.ai
@arena
X

Exciting news - DeepSeek V4 Pro is in the Arena with 1.6T parameters (49B activated) alongside V4 Flash at 284B parameters (13B activated). Both support 1M token context. It’s a major leap over DeepSeek V3.2! Code Arena: - DeepSeek V4 Pro (thinking): #3 open model (#14 overall), on par with GPT-5.4-high and Gemini-3.1-Pro in agentic webdev tasks Text Arena: - DeepSeek V4 Pro (thinking): #2 open model (#14 overall), matching Kimi-2.6 - DeepSeek V4 Flash (thinking): #10 open model (#47 overall) Competition at the top of the open model leaderboards keeps heating up. Huge congrats to @DeepSeek_AI on the strong comeback!

152retweets1.8klikes
View on X

Still wondering? A few quick answers below.

DeepSeek V4 Pro is a large-scale Mixture-of-Experts AI model featuring 1.6 trillion total parameters, with 49 billion parameters activated during any single request. It is designed for high-performance reasoning and coding tasks, supporting a standard 1-million-token context window that allows it to process massive amounts of information in a single session.

In the latest Arena.ai rankings, DeepSeek V4 Pro secured the position of the number three open-weight model for coding and number two for general text. Overall, it ranks fourteenth globally across both categories, demonstrating that it can compete directly with the world's most advanced proprietary models in human preference testing.

According to Arena.ai evaluation data, DeepSeek V4 Pro performs on par with proprietary frontier models like GPT-5.4-high and Gemini-3.1-Pro specifically in agentic web development tasks. This indicates that the open-weight model is capable of handling complex, multi-step software engineering workflows with a level of proficiency previously reserved for closed-source systems.

DeepSeek V4 Pro is the flagship model with 1.6 trillion parameters and 49 billion active parameters, while DeepSeek V4 Flash is a smaller, more efficient version with 284 billion total and 13 billion active parameters. While both support a 1-million-token context window, the Pro version ranks significantly higher on performance leaderboards for complex reasoning.

Yes, the DeepSeek V4 model family is officially open-sourced, allowing developers to download and run the models on their own infrastructure. This release provides public access to frontier-level capabilities, including the 1-million-token context window and advanced reasoning modes, without the requirement of using a proprietary API or closed-source platform.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Arena

See all AI news & updates from Arena →

Share this update