DeepSeek R2 is the most-searched AI model that does not exist. As of July 2026 there is no official announcement, no API entry, and no model card — every 'R2 launch' date that has circulated since early 2025 has come from unsourced rumor cycles, not from DeepSeek.
What actually happened instead: DeepSeek shipped the V4 family on April 24, 2026. V4 Pro integrates strong reasoning directly rather than splitting it into a separate R-series model, and community consensus reads V4 as the release that absorbed the 'R2' roadmap. Whether DeepSeek ever ships a standalone R2 — or folds reasoning into V-series releases permanently — remains unannounced.
This page tracks the official status. If R2 ships, the specs, license, and pricing will appear here; until then, treat any claimed release date as rumor. If you need DeepSeek-class open-weight reasoning today, DeepSeek V4 Pro is the current answer, with R1-0528 as the proven budget option.
Frequently asked questions
When will DeepSeek R2 be released?
There is no official release date. As of July 2026, DeepSeek has not announced R2 at all. The April 2026 DeepSeek V4 release is widely read as having absorbed the R2 roadmap by integrating reasoning into the main model line.
What should I use instead of DeepSeek R2?
DeepSeek V4 Pro is the current DeepSeek flagship for reasoning-heavy work (MIT license, 1M context). For a proven, cheaper option, DeepSeek R1-0528 and its distills remain excellent open-weight reasoning models.
Will DeepSeek R2 be open source?
If it ships, almost certainly open-weight: every DeepSeek model release to date — V2, V3, R1, V3.2, and the V4 family — has published weights under MIT.