Gemma 5, served as an API
Google's next open-weight Gemma, the line that keeps putting multimodal reasoning within reach of a single GPU. Point the OpenAI SDK at our base URL and call it per token, no contract and no minimums.
- Text and image in
- Apache 2.0 expected
- Reasoning and tool calling
- Per token, no contract
jk. It is not out yet.
Google has not announced Gemma 5, and unlike some of the models on this site there is not even a rumor mill to report. Gemma 4 landed April 2, 2026, and DeepMind has been shipping research and derivative releases since rather than a numbered successor.
Below is every rumor we could source, what Google 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 Gemma 5 rumor mill.
Each row labelled by how much it is worth. Last checked September 24, 2026.
- Release status
- ConfirmedUnreleased. Gemma 4 launched April 2, 2026 under Apache 2.0, a change from the source-available Gemma Terms used through Gemma 3. It ships in four variants: E2B and E4B for edge, a 26B MoE, and a 31B dense. Google, Gemma 4
- Leaks
- No signalNothing credible. No leak, no roadmap hint, and no analyst note naming a Gemma 5.
- Cadence math
- UnverifiedGemma February 2024, Gemma 2 June 2024, Gemma 3 March 2025, Gemma 4 April 2026. Nine to twelve months between generations puts a next one in late 2026 or 2027. That is arithmetic, not reporting.
- What DeepMind shipped instead
- ConfirmedDerivative and research work on the Gemma 4 family rather than a successor: Gemini Diffusion research, compression for mobile and laptop inference, an encoder-free multimodal model, and multi-token prediction drafters.
- License
- UnverifiedApache 2.0 expected, following Gemma 4's move off the Gemma Terms of Use. Not confirmed for any future release.
- Not Gemini
- ConfirmedGemini is Google's hosted closed line and is not licensed for third-party serving. We do not host it, this page is not about it, and no provider outside Google can offer it.
- How it would serve here
- ConfirmedSelf-hosted on our own GPUs, the same path Gemma 4 31B runs on today, including the quantized variant work.
- Price
- No signalUnknown until we have the model and can measure what it costs to serve.
What you can call right now.
The Google models live on the API today, at the rates they bill at.
Full inference catalogGemma 4 31B
The current Gemma here: dense 31B with text and image input, reasoning, tool calling, 32K context on this deployment.
Input / output per 1M tokens
$0.14 / $0.40
Gemma 4 31B NVFP4 32K
The same model quantized to NVFP4, for lower token rates on high-volume chat where the quality gap does not show.
Input / output per 1M tokens
$0.081 / $0.306
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.
We already serve two Gemma deployments
The full-precision model and an NVFP4 quantization of it both run on hardware we operate, so a new Gemma lands on a serving path that exists, including the quantized variant work.
Open weights make this a days problem
The work between a public weights drop and a served endpoint is engine support and capacity. Both are standing for this family.
Nothing to migrate when it arrives
Same base URL, same OpenAI SDK, same API key. Adopting the next Gemma 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 Gemma 5 is serving here, with the model id and the rate. Nothing else goes to this list.
Gemma 5, answered.
Is Gemma 5 out?
No. As of September 2026 Google has not announced it. Gemma 4 is the current generation, and the 31B model runs on OpenRelay today with text and image input.
Are there any Gemma 5 rumors?
None. This is the honest answer and it is why the rows above are release history and cadence arithmetic instead of leaks. Anyone publishing Gemma 5 specs today is making them up.
When is the Gemma 5 release date?
Unannounced. On the historical cadence of nine to twelve months between generations, late 2026 or 2027 is the arithmetic, but Google has not said anything. When it ships, this page becomes the live model page and the notify list gets an email.
Will OpenRelay serve Gemma 5?
Assuming it ships with open weights like every Gemma before it, yes. We self-host the family on our own GPUs, so a new release is our own engineering work rather than waiting on an upstream provider.
What should I use until then?
Gemma 4 31B for vision-grounded chat, or the NVFP4 deployment when throughput and rate matter more than the last point of quality. Both are on the cards above and take the same OpenAI SDK call.
Is this about Gemini?
No. Gemini is Google's hosted, closed line and is not licensed for third-party serving, so we do not host it. Gemma is the open-weight line, which is why it can run here at all.
More models
The account outlasts the model.
One API key, one base URL, per-token billing. Whatever Google ships next, adopting it is a one-line change rather than a new vendor.