News - Jul 3, 2026
DeepSeek R2 Goes Open Source: What It Actually Means
DeepSeek has not released R2 yet. As of mid-2025, DeepSeek R2 remains unconfirmed and unreleased — there is no official announcement, no model weights, no pricing sheet, and no changelog to report. What exists are rumors, leaks, and speculation circulating on AI forums and social media. DeepSeek R1 was released open source under an MIT license in January 2025, with weights available on Hugging Face. That release was real and significant — R1 matched or beat OpenAI o1 on several benchmarks at a fraction of the cost to run. If R2 follows the same pattern, it would be a meaningful development. But it has not happened yet. Who would benefit if R2 launches open source: developers and companies who want to self-host a frontier-grade reasoning model without paying API fees. Who gets hurt: OpenAI, Anthropic, and other closed-model providers who charge for API access. Who does not benefit right now: anyone who reads a hyped article treating rumors as news and makes decisions based on it. We will cover R2 the moment there are verifiable model weights, a license, and benchmark numbers to report.
What changed
Nothing confirmed. DeepSeek R2 has not been officially released as of the time of writing. R1 (MIT license, January 2025) remains the last verified open-source release from DeepSeek.
Who this affects
No one yet, because the release has not happened. If and when it does, self-hosting developers and cost-conscious AI teams stand to gain the most.
Our take
Publishing a news post about an unconfirmed release would be exactly the kind of hype we exist to push back against — so we are not doing it.
No official DeepSeek announcement, blog post, or model repository confirms R2 as of mid-2025. This post will be updated upon verified release.
VerdictBloom is editorially independent. No company reviewed or approved this article before publication.