DeepSeek Unveils V4-Pro AI Model to Compete in the Global AI Race
DeepSeek has released its V4-Pro AI model and updated its pricing, increasing the cost per million output tokens during peak hours from $0.87 to $3.96. Even with the price, DeepSeek is still cheaper than competitors like Anthropic’s Fable 5.
On August 13th, DeepSeek formally unveiled their V4-Pro model. Thus, following its preview period from April, the model was made available throughout the company's app, online interface, and API. The latest version, DeepSeek-V4-Pro-0813, is all about agent capabilities, according to the business. These are the parts of AI systems that handle tool usage, code execution, and multi-step workflows without any help from humans.
According to DeepSeek's benchmark findings, the model achieved impressive results on various agent-focused tests, including Terminal Bench 2.1 (87.9), DeepSWE (62.7), and NL2Repo (61.5). Up to one million tokens can be handled by the model in a context window. Users can choose between thinking and non-thinking modes, and it can generate outputs of up to 384,000 tokens.
DeepSeek Enhancing its AI Tools Portfolio
Additionally, DeepSeek has revealed that the V4-Pro API has been upgraded to natively support the OpenAI Responses API format and has Codex integration capabilities. According to the business, developers can now adjust the level of thinking effort for both V4-Pro and V4-Flash to fit the complexity of their tasks across three different settings: low, high, and max. At 6:00 UTC on August 16, a price increase will be implemented for the V4 model family.
According to DeepSeek, they are also implementing peak and off-peak invoicing, with off-peak charges being half of what they are during peak hours. Independent testing performed after the release of DeepSeek's less expensive V4-Flash variant earlier this month found that it performed above expectations, prompting the launch. Plus, there were instances where it beat the April preview of V4-Pro, which was noteworthy because Pro is marketed as the superior product by the company.
DeepSeek’s Pricing Model
In conjunction with the launch, DeepSeek announced a price hike; at peak hours, the price of V4-Pro output tokens will rise from the current flat rate of $0.87 per million to $3.96 per million. While some of DeepSeek's rivals, like Anthropic's Fable 5, charge $50 for one million output tokens, DeepSeek's pricing stays lower even at the new peak rates. From May 1st until the 5th, DeepSeek featured a sale price of 75% off V4-Pro. The price change on 13 August was thus a major shift from the company's previous approach.
For a short period in early 2025, DeepSeek held control in the artificial intelligence contest thanks to its R1 model, which went viral. But since then, rivals like Moonshot AI, Alibaba, and ByteDance have released their own products, narrowing the gap. In June, the firm raised $7.4 billion in its initial fundraising round. Plus, DeepSeek was reportedly talking to investors in July about a potential second funding round that would put its valuation at about $74 billion.
