DeepSeek R2, served as an API
The successor to DeepSeek-R1, which Reuters reported in 2025 was planned for that May. Every model here takes the same OpenAI SDK call, so if it ever serves here, adopting it is a one-string change.
- Reported: delayed since 2025
- Reasoning now a V-series mode
- R1 shipped under MIT
- Per token, no contract
jk. It is not out yet.
DeepSeek has not shipped R2, and has published no model card or API id for it. Reasoning moved into the V line instead, as a thinking mode on its general models. If you searched for a DeepSeek R2 API, nobody is serving one.
Below is every rumor we could source, what DeepSeek has actually confirmed, and the model you can call in the meantime. When the real one ships, this page becomes its page at this URL.
The DeepSeek R2 rumor mill.
Each row labelled by how much it is worth. Last checked October 4, 2026.
- Release status
- ConfirmedUnreleased. DeepSeek's last release under the R name is R1-0528, from May 28, 2025. There is no R2 model card on Hugging Face and no R2 id on DeepSeek's API. DeepSeek, R1-0528 release
- Timing
- ReportedReuters reported in early 2025 that R2 was planned for May of that year. In June 2025 it relayed The Information's report that the timing was undecided because CEO Liang Wenfeng was not satisfied with the model's performance. (Reuters, June 26, 2025)
- Dates that did not hold
- ReportedA person familiar with DeepSeek's plans denied a rumored August 15 to 30, 2025 launch, and the company had already refuted an earlier March 17 date. Neither date came from DeepSeek. TechNode, August 14, 2025
- Where reasoning went
- ConfirmedInto the V line. DeepSeek-V3.1, in August 2025, made thinking and non-thinking two modes of one model, and DeepSeek rated its thinking answers comparable to R1-0528. V4 kept both modes on every model. DeepSeek-V3.1 model card
- The R1 endpoint
- Confirmeddeepseek-reasoner, the API id R1 served under, was retired after July 24, 2026. On DeepSeek's API, reasoning is now a thinking setting on a V-series model. DeepSeek, V4 release notes
- What DeepSeek names next
- ConfirmedV4.1-Pro. DeepSeek's September 10, 2026 release notes route V4-Pro traffic to V4.1-Flash until V4.1-Pro launches. That is the next model DeepSeek's own notes name, and it is not R2. DeepSeek, V4.1-Flash release notes
- How it would serve here
- ConfirmedSelf-hosted on our own B200s, the same path DeepSeek V4-Flash runs on today, if the weights ship open as R1's did under MIT. No upstream API provider sits between us and the weights.
- Price
- No signalUnknown until we have the model and can measure what it costs to serve. Anyone quoting one is guessing.
What you can call right now.
The DeepSeek models live on the API today, at the rates they bill at.
Full inference catalogWhy it shows up here early.
Not a promise about a date. A description of the path a new open model takes to get behind this API.
DeepSeek already runs here
V4-Flash serves on B200s we operate, from NVIDIA's NVFP4 checkpoint of DeepSeek's weights. A new DeepSeek model lands on a path we have already walked.
R1 set the license precedent
R1 shipped full weights under MIT. If its successor does the same, serving it is engine work and capacity on our side, with no upstream provider to wait on.
Nothing to migrate when it arrives
Same base URL, same OpenAI SDK, same API key. Switching is a change to the model string.
One email, the day it is callable.
Leave an address and we send exactly one message when DeepSeek R2 is serving here, with the model id and the rate. Nothing else goes to this list.
DeepSeek R2, answered.
Is DeepSeek R2 out?
No. As of October 2026 DeepSeek has not released R2, and its last R-series model is R1-0528 from May 2025. DeepSeek V4-Flash, which reasons before it answers, runs on OpenRelay today.
Why was DeepSeek R2 delayed?
Reuters, relaying The Information in June 2025, reported that CEO Liang Wenfeng was not satisfied with its performance. The launch dates that circulated afterwards, including an August 2025 window, were denied.
When is the DeepSeek R2 release date?
Unannounced. The next model DeepSeek's own release notes name is V4.1-Pro, not R2. If R2 or its successor serves here, this URL becomes its page and the notify list gets one email.
Is DeepSeek R2 still coming?
Nobody outside DeepSeek knows. Every DeepSeek general model since V3.1 has a thinking mode, and the deepseek-reasoner id R1 shipped under was retired in July 2026, so a separate R2 matters less than it did in early 2025. That last part is our reading, not reporting.
Will OpenRelay serve it?
Only with open weights, because we serve models we run ourselves. R1 shipped under MIT and DeepSeek V4-Flash already runs on our GPUs, so the serving path exists.
What should I use until then?
DeepSeek V4-Flash, on the card above. It reasons before every answer and calls tools, through the standard OpenAI SDK call.
The account outlasts the model.
One API key, one base URL, per-token billing. Whatever DeepSeek ships next, adopting it is a one-line change rather than a new vendor.