DeepSeek V5, served as an API
The next DeepSeek foundation model is the release the open-weight leaderboard is waiting on, reportedly built from scratch for agentic and visual coding. Point the OpenAI SDK at our base URL and call it per token, no contract and no minimums.
- Unverified: September window
- Unverified: agents and visual coding
- Open weights expected
- Per token, no contract
jk. It is not out yet.
DeepSeek has not shipped V5. The September window doing the rounds came from community posts, not from DeepSeek: there is no model card, no API string, and no technical report. If you searched for a DeepSeek V5 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 V5 rumor mill.
Each row labelled by how much it is worth. Last checked September 24, 2026.
- Release status
- ConfirmedUnreleased. DeepSeek's newest model is V4.1-Flash, out September 10, 2026, which retired V4-Flash and its vision build on DeepSeek's API and takes V4-Pro traffic until V4.1-Pro launches. V4-Pro and V4-Flash came out April 24, 2026. DeepSeek, V4.1-Flash release notes
- Timing
- UnverifiedA September 2026 window circulated from posts in mid-to-late August, with follow-ups around September 6 claiming it was targeting the following week. None of it traces back to a DeepSeek channel.
- Capabilities
- UnverifiedPersonal agents, visual coding, UI reconstruction, and autonomous workflows, from pre-release community reports rather than anyone's hands-on evaluation.
- The leak that was not
- ConfirmedOne widely repeated “new model on July 24” leak was a misread of a routine deprecation notice: DeepSeek's API docs said deepseek-chat and deepseek-reasoner would be discontinued that day. Worth knowing before you trust the next screenshot.
- The case against
- UnverifiedV4 is only a few months old, and DeepSeek's rhythm makes a V4.x post-training pass or another V4 variant at least as likely as a new generation.
- The case for
- ConfirmedDeepSeek is quiet before launches. V3.1 and V4-Flash both appeared with very little warning, so no pre-announcement is not evidence of nothing coming.
- Where it will actually appear
- ConfirmedA real release shows up on api-docs.deepseek.com, DeepSeek's own account, or a Hugging Face model card. If api.deepseek.com serves a deepseek-v5 id, it exists. Everything before that is someone's screenshot.
- How it would serve here
- ConfirmedSelf-hosted on our own GPUs. 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.
These models are live on the API today, at the rates they bill at.
Full inference catalogGPT-OSS 120B
A general model for reasoning and agents: OpenAI's 120B open-weight model, with tool calling and structured output.
Input / output per 1M tokens
$0.15 / $0.60
DeepSeek-OCR 2
The same lab's OCR model, for turning scans, invoices, and PDFs into markdown or JSON at per-token rates.
Input / output per 1M tokens
$0.039 / $0.039
Why 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 has run here in production
DeepSeek V4-Flash ran on hardware we operate, from NVIDIA's NVFP4 checkpoint of DeepSeek's weights. Standing up a new DeepSeek mixture-of-experts is a path we have walked rather than a first attempt.
Open weights make this a days problem
The work between a Hugging Face release and a served endpoint is engine support and capacity, and we hold both standing for this family. There is no partner to negotiate with.
Nothing to migrate when it arrives
Same base URL, same OpenAI SDK, same API key. Adopting the next DeepSeek means changing the model string in one line.
One email, the day it is callable.
Leave an address and we send exactly one message when DeepSeek V5 is serving here, with the model id and the rate. Nothing else goes to this list.
DeepSeek V5, answered.
Is DeepSeek V5 out?
No. As of September 2026 DeepSeek has not released it and there is no official announcement. GPT-OSS 120B and DeepSeek-OCR 2, on the cards above, run here today.
Is the September release date real?
It is not from DeepSeek. The date came from community posts in August and early September 2026, with no model card, API string, or technical report behind it. Treat it as speculation until api.deepseek.com or a Hugging Face card says otherwise.
When is the DeepSeek V5 release date?
Unannounced. DeepSeek does not publish a forward roadmap and has a habit of shipping with little warning, which is the whole reason this page is worth being on. When it lands, this page becomes the live model page and the notify list gets an email.
Will OpenRelay serve DeepSeek V5?
Assuming it ships with open weights like every DeepSeek model before it, yes. We would self-host it on our own GPUs, as we did DeepSeek V4-Flash, rather than resell another provider's endpoint, so a new release is our own engineering work rather than a partnership.
What should I use until then?
GPT-OSS 120B for reasoning, agents, and tool calling, or DeepSeek-OCR 2 if the job is documents. Both are on the cards above and both take the same OpenAI SDK call.
Why does this page exist if the model does not?
Because people search for the next model before it ships, and landing on nothing is worse than landing on a sourced answer. Every rumor above is labelled by what it is worth, the model you can actually call is one click away, and this URL becomes the real model page at launch.
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.