Anything you can do I can do better…
That may be the motto of the AI arms race, which unfolds in many places in 2026. There is competition between Silicon Valley AI labs such as Anthropic, OpenAI, and Google DeepMind, a race for chips and computing power, and, of course, a fierce competition between the US and China.
In the recent battle of East vs. West in the AI battle, Chinese artificial intelligence company DeepSeek has released a preview of its latest model, the DeepSeek V4.
Here’s what we know about the new model, and how it stacks up against other new models like OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.7.
Want to learn more about getting the most out of your technology? Sign up for Mashable’s Top Stories and Deals newsletters today.
OpenAI’s GPT-5.5 vs Claude Opus 4.7: Which is better?
What is DeepSeek V4?
DeepSeek V4 Preview is a new open source AI model. And that’s the big difference between DeepSeek and its competitors from US companies — a true open source model.
Anthropic, OpenAI, xAI, and other US companies strictly protect their models at the border, while DeepSeek is available for anyone to download and modify under the MIT license.
Two versions of the new open source model are available for use. Users can also access new models via chat.deepseek.com.
Mashable Light Speed
DeepSeek strengthens China’s edge in open source AI.
A few days before the launch of DeepSeek V4, another Chinese AI company, Moonshot AI, released the open source model Kimi K2.6, although early testers say that DeepSeek V4 is clearly a better model. A full system card is available with Hugging Face.
AI is creating a zero-day bug detection problem, and it’s getting worse
DeepSeek V4: Working with benchmarks
DeepSeek says its new open source model is a big step forward in agent operations and coding. The company also says that “DeepSeek-V4 is seamlessly integrated with leading AI agents such as Claude Code, OpenClaw and OpenCode.”
DeepSeek has released benchmark results showing that the new models achieve the same results as recent models from OpenAI, Google, and Anthropic.
DeepSeek V4 benchmark performance comparison.
Credit: DeepSeek
On popular AI leaderboards like Arena and synthetic analytics, DeepSeek V4 currently lags behind the top-tier models, though that could change in the coming days and weeks.
DeepSeek V4 is more economical than the competition
You may remember when DeepSeek burst onto the AI scene on Jan. 2025, when the company announced its R1 model. That model sent shockwaves through the US AI industry by proving that it is possible to build powerful models more efficiently and cheaply.
DeepSeek V4 is also much cheaper than Gemini, ChatGPT, and Anthropic. That could give DeepSeek a big edge in discovery.
Here’s how API pricing compares:
-
DeepSeek V4 it costs $1.74 for 1 million input tokens and $3.48 for 1 million output tokens (1 million content window)
-
GPT-5.5 it costs $5 for 1 million input tokens and $30 for 1 million output tokens (1 million context window)
-
Claude Opus 4.7 it costs $5 for 1 million input tokens and $25 for 1 million output tokens (1 million content window)
-
Google Gemini 3.1 Pro it costs $2 for 1 million input tokens and $12 for 1 million output tokens
As you can see, the DeepSeek is about a sixth of the cost of the latest US models, which is a huge plus. Even the more affordable Gemini 3.1 Pro is in dire straits.
So, a job that would cost you $5.22 with DeepSeek V4 will cost you $35 with GPT-5.5, or about 85 percent less.
Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging that it infringes Ziff Davis’s copyright in training and using its AI programs.