DeepSeek V4 is here: How it compares to ChatGPT, Claude, Gemini
DeepSeek V4 is a new open-source AI model released by DeepSeek, a Chinese AI company. It is notable for being open-source under an MIT license, unlike U.S. competitors like Anthropic, OpenAI, and Google DeepMind, which tightly control their models.
Two versions of DeepSeek V4 are available: DeepSeek-V4-Pro (1.6 trillion parameters) and DeepSeek-V4-Flash (284 billion parameters). The model is accessible via chat.deepseek.com and Hugging Face. Early tests suggest DeepSeek V4 outperforms Moonshot AI’s Kimi K2.6.
DeepSeek V4 excels in agentic tasks and coding, with seamless integration with leading AI agents like Claude Code, OpenClaw, and OpenCode. Benchmark results indicate it performs comparably to the latest frontier models from OpenAI, Google, and Anthropic, though it currently lags on popular AI leaderboards like Arena and Artificial Analysis.
Cost-wise, DeepSeek V4 is significantly cheaper than competitors: $1.74 per 1 million input tokens and $3.48 per 1 million output tokens. This is about one-sixth the cost of GPT-5.5 ($5/$30), Claude Opus 4.7 ($5/$25), and even cheaper than Google’s Gemini 3.1 Pro ($2/$12). For example, a task costing $5.22 with DeepSeek V4 would cost $35 with GPT-5.5, representing an 85% cost reduction.
DeepSeek’s R1 model from January 2025 already demonstrated efficiency and cost-effectiveness, reinforcing its position in the open-source AI space. The company is reinforcing China’s lead in open-source AI development.
Source: Mashable