China’s DeepSeek Releases New AI Models It Claims ‘Nearly’ Match Up To OpenAI, Anthropic Tech

• DeepSeek’s new V4-Pro model trails only Google’s Gemini 3.1-Pro for world knowledge and the overall performance is only “marginally short” of OpenAI’s GPT‑5.4, the company said.
• The V4-Pro delivered better results than Anthropic’s Sonnet 4.5 but came in behind the U.S. startup’s flagship Opus 4.6 model.
• DeepSeek added that the V4 series has been tested on Huawei’s Ascend NPUs, marking a milestone for Chinese chips.

DeepSeek, one of the most important and closely watched AI startups from China, on Friday released its latest AI models and touted its ability to go toe-to-toe with AI from U.S. rivals such as OpenAI, Anthropic and Google.

The release comes after a long period of lull and over a year since its then-flagship model, the DeepSeek-R1, sent shockwaves through the global tech scene by matching benchmark capacities while being developed at a fraction of the cost of Western models.

In the WeChat release, DeepSeek said its V4-Pro and V4-Flash were nearly as proficient as the top-of-the-line AI models in the market.

The Pro model beats all rival open models for maths and coding, and trails only Google’s Gemini 3.1-Pro for world knowledge, DeepSeek said. Its performance is “marginally short” of OpenAI’s GPT‑5.4 model.

It beats Anthropic’s Sonnet 4.5 in user experience and delivers output quality closer to Opus 4.6’s non-thought mode, though it still lags behind Opus 4.6’s thought mode, the company said.

DeepSeek’s V4 Runs On Huawei Chips

DeepSeek also said it has successfully tested one of V4’s core efficiency methods across both Nvidia GPUs and Huawei’s Ascend NPUs, marking a milestone for Chinese chips. 

In a WeChat update, Huawei added that its Ascend platform now provides full-stack compatibility for DeepSeek V4 models.

While DeepSeek has repeatedly delayed releasing a major model update, domestic tech giants Alibaba and ByteDance have aggressively pushed out updates.

Alibaba, which has pushed updates each month of the year, is rapidly pivoting Qwen toward agentic AI, multimodal capabilities, and mass consumer adoption.

OpenAI, Anthropic Prep For IPO

The development comes as OpenAI and Anthropic prepare for their initial public offerings (IPOs), likely to arrive in the final quarter of 2026. Anthropic recently disclosed that its annual revenue run rate had hit $30 billion, surpassing that of OpenAI.

Both firms are yet to file their paperwork. However, retail investors are already gaining exposure to these marquee pre-IPO tech names through funds such as ARK Innovation (ARKK), Fundrise Innovation Fund (VCX), Destiny Tech100 (DXYZ), and KraneShares Artificial Intelligence & Technology ETF (AGIX).

Ark Invest recently picked up a $240 million stake in OpenAI through its funds ARKK, ARK Fintech Innovation (ARKF), and ARK Next Generation Internet (ARKW). Stocktwits previously published explainers on investing in Anthropic and SpaceX ahead of their IPOs.

For updates and corrections, email newsroom[at]stocktwits[dot]com.

Read Next: POET Stock Eyes Best Week In Nearly 2 Years: Traders Now Hint At A Possible Amazon Connection