Chinese AI startup DeepSeek has unveiled its latest innovation, the V3.1 large language model, revolutionizing the landscape of open AI with its groundbreaking capabilities. This model, with a staggering 685-billion parameters, competes head-to-head with proprietary systems from major industry players like OpenAI and Anthropic but with a significant cost advantage. Notably, V3.1 achieves an impressive 71.6% on coding benchmarks, signaling its robust performance in real-world applications.
One of the distinguishing features of DeepSeek V3.1 is its hybrid architecture, seamlessly integrating chat, reasoning, and coding functionalities within a unified framework. This architectural evolution is augmented by its hybrid inference capabilities, introducing “Think” and “Non-Think” modes. These modes offer users enhanced reasoning speed and improved agent functionalities, pushing the boundaries of traditional AI capabilities.
The release includes substantial updates, such as support for a 128,000 token context, ensuring comprehensive and extended text processing. The updated model is compatible with the Anthropic API format, includes strict function calling in beta, and provides API updates that cater to various practical applications.
From a technical perspective, V3.1 demonstrates optimized performance for new Chinese chips, utilizing the UE8M0 FP8 data type. This progression hints at the potential of forthcoming domestic silicon from Chinese tech giants like Huawei, given the current limitations of chips like the Ascend 910C in supporting FP8 natively.
Available for free on platforms like Hugging Face, the model challenges conventional AI business strategies by delivering frontier-level capabilities without imposing usage restrictions or licensing fees. Alongside the open-source availability of weights, the model sets a new benchmark in accessible AI technology.
The revised pricing model for the API will be effective from September 5th, 2025, but the focus remains on affordability and expansive access. DeepSeek V3.1 represents a significant leap in AI technology, effectively democratizing access to cutting-edge advancements and heralding a new era of AI accessibility and capability.

Collections

DeepSeek has released V3.1, a 685-billion parameter open-source language model that rivals proprietary systems from OpenAI and Anthropic at significantly lower costs. The model features a hybrid architecture combining chat, reasoning, and coding capabilities, supports 128,000 token context, and achieves 71.6% on coding benchmarks. Available for free on Hugging Face with API compatibility, it's optimized for Chinese chips and represents a major step toward democratizing advanced AI technology.

DeepSeek-V3.1 Release: A New Era in Open AI Technology

How does it compare to the newest open AI ?

Another day, another era. From the Kiro IDE to GPT-5 now DeepSeek. I’ll bet in two weeks there’ll be more so called “revolutionary” stuff to come.

DeepSeek is just CCP propaganda dressed up as AI. It censors Taiwan, rewrites CCP history, and justifies China’s territorial grabs. Didn’t OpenAI already debunk it as a hacked GPT clone?

Way to promote Communism that killed millions…because AI.