DeepSeek releases V4 model preview with agent optimization, pricing undisclosed
DeepSeek released a preview of its V4 large language model on April 24, 2026, available in 'pro' and 'flash' versions. The Hangzhou-based company claims the open-source model achieves strong performance on agent-based tasks and has been optimized for tools like Anthropic's Claude Code and OpenClaw.
DeepSeek released a preview version of its V4 large language model on April 24, 2026, marking the company's first major model update since its R1 reasoning model in January 2025. The model is available in two variants: "pro" and "flash," though the company has not disclosed technical specifications, pricing, or context window sizes.
Model Capabilities
According to DeepSeek, V4 delivers improved performance against domestic Chinese competitors, particularly in agent-based tasks, knowledge processing, and inference. The company specifically optimized the model for compatibility with popular agent tools including Anthropic's Claude Code and OpenClaw.
Like its predecessor V3, DeepSeek-V4 is open source, allowing developers to download the code, run it locally, and modify it. The company has not released benchmark scores or comparative performance data.
Market Context
The release comes 15 months after DeepSeek's R1 reasoning model disrupted global tech markets in January 2025. R1 matched or outperformed leading models from OpenAI and Google on several benchmarks, despite DeepSeek's claims of development costs far below U.S. competitors.
DeepSeek's V3 model, released in late 2024, gained attention for reportedly being trained with less powerful chips and at a fraction of the cost of comparable models. However, the company's subsequent model releases have not replicated R1's market impact.
Competitive Landscape
DeepSeek now faces intensifying competition in China's AI sector. Alibaba and ByteDance have both released new models in 2026, competing for market share in the rapidly growing domestic AI market.
Founded in 2023 and based in Hangzhou, DeepSeek continues its strategy of open-source releases, contrasting with the closed-source approaches of many Western AI labs.
What This Means
The V4 preview extends DeepSeek's open-source model lineup but lacks the specificity needed to assess its competitive position. Without disclosed benchmarks, pricing, or technical specifications, it's unclear whether V4 represents a meaningful advance over V3 or how it compares to recent releases from competitors like Alibaba's Qwen and international models. The focus on agent optimization suggests DeepSeek is targeting enterprise and developer use cases, though the absence of pricing information makes cost comparisons to Western alternatives impossible.
Related Articles
DeepSeek V4 Pro launches with 1.6T parameters at $1.74/M tokens, undercutting Claude Sonnet 4.6 by 42%
DeepSeek released two preview models: V4 Pro (1.6T total parameters, 49B active) and V4 Flash (284B total, 13B active), both with 1 million token context windows. V4 Pro is priced at $1.74/M input tokens and $3.48/M output—42% cheaper than Claude Sonnet 4.6—while V4 Flash at $0.14/$0.28 per million tokens undercuts all small frontier models.
DeepSeek Releases V4-Flash-Base: 292B Parameter Base Model
DeepSeek has released V4-Flash-Base, a 292 billion parameter base model now available on Hugging Face. The model uses BF16, I64, F32, and F8_E4M3 tensor types and is distributed in Safetensors format.
DeepSeek Releases V4-Pro-Base with 1.6 Trillion Parameters
DeepSeek has released DeepSeek-V4-Pro-Base, a 1.6 trillion parameter foundation model now available on Hugging Face. The base model uses BF16 precision for weights and includes support for F8_E4M3, I64, and F32 tensor types.
DeepSeek Releases V4-Flash: 284B-Parameter MoE Model With 1M Token Context at 27% Inference Cost
DeepSeek released two Mixture-of-Experts models: V4-Flash with 284B total parameters (13B activated) and V4-Pro with 1.6T parameters (49B activated). Both models support one million token context windows and use a hybrid attention architecture that requires only 27% of the inference FLOPs compared to DeepSeek-V3.2 at 1M token context.
Comments
Loading...