Overview
- DeepSeek's public API 'Models & Pricing' page now lists a new model called DeepSeek‑V4‑Pro‑0813 and shows support for 'thinking' and 'non‑thinking' operational modes.
- The entry appeared on DeepSeek's official documentation as reported by IT之家 but DeepSeek has not published a changelog or formal release notes for the 0813 entry.
- Earlier this month DeepSeek made its V4‑Flash variant available for public beta, and the new documentation entry follows the company's prior statement that a V4‑Pro release would arrive 'as soon as possible.'
- The labels 'thinking' and 'non‑thinking' likely refer to different inference or behavior settings for the model, but the documentation does not define those terms or explain technical differences.
- Community tipsters tipped off reporters to the change, and the main next steps for users are to wait for DeepSeek's official changelog and availability notice to learn exact capabilities, pricing and access changes.