Back to Blog
April 24, 2026 • Ram Iyer

DeepSeek V4: Closing the Gap with Frontier AI Models

DeepSeek V4: Closing the Gap with Frontier AI Models DeepSeek V4: Closing the Gap with Frontier AI Models DeepSeek V4: Closing the Gap with Frontier AI Models

DeepSeek Previews New AI Model That ‘Closes the Gap’ with Frontier Models

Chinese AI lab DeepSeek has launched two preview versions of its newest large language model, DeepSeek V4. This update to last year’s V3.2 model and its accompanying R1 reasoning model aims to significantly close the performance gap with leading AI models.

Key Features of DeepSeek V4

  • Mixture-of-Experts (MoE) Architecture: Both DeepSeek V4 Flash and V4 Pro utilize an MoE approach. This architecture activates only a subset of parameters for each task, leading to lower inference costs and improved efficiency.
  • Massive Context Windows: Each model boasts a context window of 1 million tokens, enabling the processing of large codebases or extensive documents within prompts.
  • Parameter Count:
    • V4 Pro: Features 1.6 trillion total parameters with 49 billion active parameters, making it the largest open-weight model currently available. This surpasses models like Moonshot AI’s Kimi K 2.6 (1.1 trillion) and MiniMax’s M1 (456 billion).
    • V4 Flash: A smaller variant with 284 billion total parameters and 13 billion active parameters.
  • Performance Gains: DeepSeek claims both V4 models are more efficient and performant than V3.2 due to architectural improvements. They are reported to have nearly closed the gap with current leading open and closed-source models on reasoning benchmarks.
  • Competitive Benchmarking:
    • The V4-Pro-Max model is said to outperform open-source peers on reasoning benchmarks and surpass OpenAI’s GPT-5.2 and Gemini 3.0 Pro on certain tasks.
    • In coding benchmarks, both V4 models are described as performing comparably to GPT-5.4.
  • Knowledge Lag: The models appear to trail frontier models like OpenAI’s GPT-5.4 and Google’s Gemini 3.1 Pro in knowledge tests, suggesting a developmental trajectory approximately 3 to 6 months behind the state-of-the-art.
  • Text-Only Support: Unlike some closed-source competitors, both V4 models currently support only text input and output.

Affordability and Competitive Pricing

DeepSeek V4 models are positioned as significantly more affordable than current frontier models:

  • V4 Flash: Priced at $0.14 per million input tokens and $0.28 per million output tokens, undercutting models like GPT-5.4 Nano, Gemini 3.1 Flash, GPT-5.4 Mini, and Claude Haiku 4.5.
  • V4 Pro: Priced at $0.145 per million input tokens and $3.48 per million output tokens, undercutting Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.7, and GPT-5.4.

Context and Accusations

The launch occurs amidst ongoing accusations against Chinese AI labs, including DeepSeek, for alleged intellectual property theft and model distillation from Western AI companies like Anthropic and OpenAI.