DeepSeek releases V4 Pro flagship model and open-source agent framework

DeepSeek V4 Pro AI agents Mixture-of-Experts API pricing open source AI

Chinese AI startup DeepSeek has officially released the general availability (GA) version of its flagship model, DeepSeek-V4-Pro (build 0813). The new model is designed to significantly enhance AI agent capabilities, particularly for complex coding and automated workflows. Built on a Mixture-of-Experts (MoE) architecture with approximately 1.6 trillion total parameters, the model supports a 1-million-token context window and features improved performance on benchmarks such as Terminal Bench 2.1, where it scored 87.9.

Alongside the model release, DeepSeek introduced DeepSeek Harness v0.1, an open-source agent framework intended to compete with developer tools like Anthropic’s Claude Code. The framework is modular, allowing developers to use various plugins for different tasks.

To manage resources, DeepSeek is transitioning from flat API pricing to a peak and off-peak pricing structure, effective mid-August. While the company is increasing its overall API rates, the off-peak rates remain highly competitive. This strategic shift comes as DeepSeek seeks to expand its workforce and computing capacity while pursuing a potential multi-billion dollar valuation in upcoming fundraising rounds.

Entities

Alibaba · Anthropic · ByteDance · Claude · DeepSeek · Hangzhou · OpenAI

Claims

What the coverage asserts, and how well corroborated each claim is across sources.

[● 3 SOURCES] The new model provides native support for the OpenAI Responses API format. www.hwupgrade.it · www.blocktempo.com · venturebeat.com
[● 2 SOURCES] DeepSeek is planning a new fundraising round with a valuation of approximately $74 billion. olhardigital.com.br · kfgo.com
[DISPUTED] The V4 Pro model utilizes a Mixture-of-Experts (MoE) architecture with approximately 1.6 trillion total parameters. ihal.it · www.blocktempo.com
[● 10 SOURCES] DeepSeek has officially released the V4 Pro flagship model (build 0813) for general availability. www.iplaysoft.com · www.hwupgrade.it · olhardigital.com.br · ihal.it · www.eeo.com.cn · +5 more
[● 4 SOURCES] DeepSeek V4 Pro achieved a score of 87.9 on the Terminal Bench 2.1 benchmark. www.iplaysoft.com · www.eeo.com.cn · tech-noisy.com · www.fonearena.com
[● 4 SOURCES] The company is introducing a peak and off-peak pricing model for its APIs, effective August 16 or 17. technews.tw · news.cnyes.com · venturebeat.com · www.blocktempo.com
[● 3 SOURCES] The V4 Pro model supports a context window of up to 1 million tokens. www.hwupgrade.it · ihal.it · www.blocktempo.com
[● 2 SOURCES] DeepSeek launched DeepSeek Harness v0.1, an open-source agent framework designed to compete with tools like Claude Code. news.cnyes.com · venturebeat.com