DeepSeek V4 is here: How it compares to ChatGPT, Claude, Gemini
Source: Mashable Tech
Anything you can do I can do better…
That may as well be the motto for the AI arms race, which is unfolding across multiple dimensions in 2026. There’s the competition between Silicon Valley AI labs like Anthropic, OpenAI, and Google DeepMind, the race for chips and compute power, and, of course, the fierce competition between the U.S. and China.
In the latest salvo in the East vs. West battle for AI supremacy, Chinese artificial intelligence company DeepSeek has released a preview of its latest model, DeepSeek V4.
What is DeepSeek V4?
DeepSeek V4 Preview is a new open‑source AI model. That’s the biggest difference between DeepSeek and its rivals from U.S. companies—it’s a true open‑source model.
Anthropic, OpenAI, xAI, and other U.S. companies fiercely protect their frontier models, whereas DeepSeek is available for anyone to download and modify under an MIT license.
Two versions of the new open‑source model are available for use. Users can also access the new models via chat.deepseek.com.
- DeepSeek‑V4‑Pro: a 1.6‑trillion‑parameter model
- DeepSeek‑V4‑Flash: a 284‑billion‑parameter model
Just days before the launch of DeepSeek V4, another Chinese AI company, Moonshot AI, released the open‑source model Kimi K2.6, though early testers say DeepSeek V4 is clearly the better model. The full system card is available via Hugging Face.
DeepSeek V4: Performance and benchmarks
DeepSeek says its newest open model makes great strides in agentic tasks and coding. The company also notes that “DeepSeek‑V4 is seamlessly integrated with leading AI agents like Claude Code, OpenClaw & OpenCode.”
Benchmark results indicate the new models achieve results comparable to the latest frontier models from OpenAI, Google, and Anthropic.
DeepSeek V4 benchmark performance comparison. Credit: DeepSeek
On popular AI leaderboards such as Arena and Artificial Analysis, DeepSeek V4 currently lags behind the top frontier models, though that could change in the days and weeks ahead.
DeepSeek V4 is far more economical than the competition
You may remember when DeepSeek burst onto the AI scene in Jan. 2025 with its R1 model, proving it was possible to build powerful models much more efficiently and cheaply.
DeepSeek V4 is also priced far lower than Gemini, ChatGPT, and Anthropic, which could give it a huge edge in adoption.
API pricing comparison (per 1 million tokens, 1 million‑token context window)
| Model | Input token cost | Output token cost |
|---|---|---|
| DeepSeek V4 | $1.74 | $3.48 |
| GPT‑5.5 | $5.00 | $30.00 |
| Claude Opus 4.7 | $5.00 | $25.00 |
| Google Gemini 3.1 Pro | $2.00 | $12.00 |
As a result, DeepSeek is about one‑sixth the cost of the latest U.S. models. A task that would cost $5.22 with DeepSeek V4 would cost roughly $35 with GPT‑5.5—about 85 % less.
Disclosure: Ziff Davis, Mashable’s parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.