AIBRIEF
DeepSeek's August 13 update of its 1.
A new DeepSeek Pro-tier model, V4 Pro 0813, surfaced on OpenRouter this week as an API-only listing with no DeepSeek announcement page behind it. OpenRouter is a third-party API aggregator that routes requests to multiple model providers; this build currently runs on a single host with 100% uptime over the past three days.
The 0813 build is a refinement of the base DeepSeek V4 Pro, which DeepSeek released on April 24, 2026. Architecture is a Mixture-of-Experts, a model that activates only a subset of its parameters per token to cut compute cost, at 1.6T total parameters and 49B activated, with a 1M-token context window and hybrid attention. OpenRouter lists the 0813 build at $0.435 per 1M input tokens and $0.87 per 1M output tokens, with up to 384,000 output tokens and reasoning-effort levels of low, medium, and high.
The benchmarks that named the build appear to have traveled through an Official DeepSeek WeChat group, a Reddit post moderators deleted as "low-effort", and a Hacker News ASCII-art table. Simon Willison's note on the 0813 listing flags the friction path: no press release in the chain. Willison's curiosity hook: the same prompt rendered visibly different pelicans at low, medium, and high reasoning effort, an unusual gap inside a single release.
Open weights are likely given DeepSeek's prior patterns but unconfirmed. Top OpenRouter traffic for the 0813 build currently comes from SillyTavern, Janitor AI, and Open WebUI.