DeepSeek V4 Pro 0813 has been released on OpenRouter, generating significant discussion on Hacker News with 84 points and 13 comments. The model represents a substantial upgrade over previous versions, offering improved reasoning, larger context window, and enhanced coding abilities. Early benchmarks suggest it competes closely with leading proprietary models while maintaining competitive pricing per token.

Users on Hacker News have noted its strong performance on complex math and programming tasks, with some comparing it favorably to Claude 3.5 Sonnet and GPT-4o. The model is accessible through OpenRouter's API, making it easy for developers to integrate. Pricing details have not been fully disclosed, but the community expects it to be cost-effective given DeepSeek's history of aggressive pricing.

Key features include a 128K context window, support for multiple languages, and improved instruction following. The model is optimized for both chat and completion tasks. Some users have reported minor issues with consistency on very long contexts, but overall sentiment is positive. The release leverages DeepSeek's proprietary architecture and training techniques.

For those interested in trying the model, the direct link is provided on OpenRouter. The Hacker News thread includes additional user experiences and performance comparisons. DeepSeek continues to push the boundaries of open-weight AI models, and V4 Pro 0813 marks a significant milestone in their development roadmap.