Update: Gemini 3.1 Rollout is Underway! 🚀

[srithreepo]

Update: Gemini 3.1 Rollout is Underway! 🚀

As many of you noticed, we officially shipped support for Gemini 3.1 yesterday! We are currently in the process of transitioning our user base to the new model. Here is the current status of the rollout:

Current Availability

For users logging in with Google Login:

• Full Access: Gemini 3.1 is now live for all AI Ultra ✅ and AI Ultra for Business ✅ users.
• Partial Rollout: A percentage of all other users now have access, with more being added as capacity permits.

For those that use API keys:

• As long as your API Key is able to access Gemini 3.1 Pro you should be able to use Gemini 3.1 Pro on Gemini CLI as well. Check here for (AI Studio)(Vertex AI)

What to Expect During the Transition

While we work to make 3.1 available to everyone as quickly as possible, you may experience a "mixed" environment. For a short window, you might see Gemini 3.1 Pro in one session and Gemini 3 Pro in another as the deployment propagates.

Verifying if you have access to Gemini 3.1 Pro

Important Notes

• Model flag: If you are using the -m flag to set Gemini 3.1 Pro, it may not function consistently until the rollout is 100% complete for your account.
• Google Login vs. API: This phased rollout specifically affects users logging in directly via the Login with Google option. If you are using an API key, you should have access to Gemini 3.1 immediately, provided your key has the necessary permissions.

We appreciate your support and patience as we scale up. Stay tuned for further updates!

[ByteX420]

[BayuuPermana]

I seen the dev said next week

[mrowlinson]

Where did you see this? I pay for Pro and have been stuck with these old ass models for days.

[Xoronics]

This is great news, but can the team provide any clarification on usage limits? I see sessions are now on a rolling 24h cycle instead of 5h (AI Ultra Workspace) and it seems to "cost" roughly 1.5% of the daily limit for a decently complex task? Is this is just temporary given "launch" capacity constraints? Spent the day hammering antigravity with 3.1 and hardly used 15% of the 5h Quota at any given time so just a little unsure about this mismatch.

[Xoronics]

oof

[mattsilv]

Is it true if I comment here I get access right away? And can you confirm anyone who reacts to this message gets access too?

[ameer-pk]

We need Opus 4.6 level intelligence but cheaper.

[SchulteDev]

Have you tried Sonnet 4.6?

[sascharo]

Have you tried Sonnet 4.6?

I did; it's shallow.

[catursurapam]

tested it but not sure the result is same as opn ai and clude

[anthonylhowardjr]

We need quicker, smoother rollouts for all tiers. As a paying customer, Gemini CLI should never lag behind Copilot (or others) on model access in the CLI.

[Jaylonnet]

We need quicker, smoother rollouts for all tiers. As a paying customer, Gemini CLI should never lag behind Copilot (or others) on model access in the CLI.

This.

[Godzilla675]

Yup

[mdgld]

Honestly hilarious that I can access 3.1 in every other tool I use—raycast, copilot, openrouter...the list goes on—and not a google product that I pay for.

[ihpled]

I don't understand why all Pro users have access to Gemini 3.1 Pro in Antigravity, while no one has access via the Gemini CLI. Is Gemini CLI already dead in favor of Antigravity?

[pasha-vuiko]

To me cli looks more like an experiment than a real tool. From my understanding it is stateless and for sure it painful slow compared to Antigravity. If I point Antigravity to the files it is very fast and efficient

I'm not agree that the CLI is more like an experiment, it is a separate tool which is handy in some specific situations, also there are AI CLIs which work fine, such as Codex even though Codex CLI was introduced much later than Gemini CLI, so I really confused why quality of Gemini CLI is quite low

[postadelmaga]

Let me try to explain

• 1 Google cli is most in experiment,imho, you can see it from "the joke talks" it spit her and there.
• 2 cli tools in general are not meant for real development because human cannot work in big project from a cli perspective (at least most of the people don't use these approach in real work environment)
  if you want to work on a big project with cli tool you can do it but you need a really good ai and we don't have it right now
  My 2 cents

[ihpled]

Why do you prefer Cli to Antigravity ? Looks to me Antigravity is much faster

Some people may prefer IDEs not based on VS Code

Yes, this is the main reason

[cacarr-pdxweb]

Let me try to explain

• 1 Google cli is most in experiment,imho, you can see it from "the joke talks" it spit her and there.
• 2 cli tools in general are not meant for real development because human cannot work in big project from a cli perspective ....
  if you want to work on a big project with cli tool you can do it but you need a really good ai and we don't have it right now
  My 2 cents

Many, many developers work in a CLI.

[postadelmaga]

Really ? Find me a big project where developer works in cli ... Sys admin works in cli, developer uses cli. Hobbyist may use cli

[mitchross]

This is why anthropic will win. All this gate keeping for paying customers is so ridiculous. Dont bother with a big fancy press release if you dont even release it to paying customers.

[Jaylonnet]

100%. They rush to release something to not get left in the dust, when all it does is irritate consumers. We'd rather wait and have a functioning product, instead of be a paying customer and not have access to something that they want to boast about.

[machinae]

Gemini CLI has become completely unusable since this update.

I am constantly getting API Error: No capacity available for model gemini-3-flash-preview on the server] errors.

Bye gemini, I'm switching to codex now. What a terrible UX.

[mazar]

Gemini 3.0 was "released" 3 months ago and as a paying customer of Gemini Code Assist Enterprise I don't have access to it and still stuck at 2.5. Seriously considering Claude instead.

[sascharo]

Gemini 3.0 was "released" 3 months ago and as a paying customer of Gemini Code Assist Enterprise I don't have access to it and still stuck at 2.5. Seriously considering Claude instead.

Am I reading this right? Gemini Code Assist Enterprise has no access to 3.0? If a product only offers Gemini 2.5, I would definitely cancel it.

[mazar]

3.0 is in preview and enterprises do not enable preview for their employees

[obey24com]

Let's see. I found that Gemini is the most hallucinating model compared to Opus or Codex, which is why I only use it for more creative parts and plans, less than actual working on the code.

[cacarr-pdxweb]

Your opinion regarding the general merits of Google's models is irrelevant to the issue of the rollout of Gemini 3.1 to Gemini CLI.

This is GitHub, not X or Reddit.

[obey24com]

It's hallucinating. That's a real issue. Please keep your personal opinions to yourself. This isn’t the place to play sheriff.

This is GitHub, not X or Reddit.

[cacarr-pdxweb]

It's hallucinating. That's a real issue. Please keep your personal opinions to yourself. This isn’t the place to play sheriff.

This is GitHub, not X or Reddit.

Go make general complaints about the model in the appropriate place. This is a discussion about the rollout of 3.1 to Gemini CLI.

[domjesus]

I am a subscriber of Google AI pro, which should refresh quota every 5hr, however last Friday I got suspended (while using 3.1 pro on Antigravity) till Monday (almost 3 days). I opened a ticket, but the support team couldn’t solve the issue yet.

[partarstu]

@domjesus Same here

[sascharo]

Why did you get suspended? Not because you used 3.1, or did you?

[partarstu]

@sascharo I suspect exactly because I used 3.1 quite intensively, 5-hour quota 3 times in 24h - because as already mentioned by many, 3.1 consumes the tokens almost 2 times faster than 3. So after my 3rd 5-hour quota was exceeded, I got a lockout for 90 hours, just like that. I always thought that weekly limits was designed for people using Claude models (never hit it with Gemini 3), but seems like 3.1 has the similar approach as with Claude models, no idea why.

[raky291]

I'm a Google AI Pro subscriber and I still don't have the 3.1 Pro model on Gemini CLI 😐

[sascharo]

I'm a Google AI Pro subscriber and I still don't have the 3.1 Pro model on Gemini CLI 😐

Same here. It feels like an eternity in our day and age. 😅

[pasha-vuiko]

that's very unfortunate to see that Google makes such really smart models as Gemini 3/3.1 and how the company fails to make it work well in its own products such as Gemini CLI and Gemini Code Assist plugin in IDEs, and now I'm even not taking about the fact that we need to wait so long to get access to the models in Google's tools when it is available from day 1 in third-party products

[lojic]

@srithreepo what is the rough ETA for when 3.1 will be available for all paid users in gemini-cli ?

I'm sure you are all acutely aware of this by now, but this 3.1 rollout was ridiculously bad. It's been over 4 days now, and as a paying customer, I still have no access. Why announce it as available in the CLI when it is not available?

[sascharo]

Hopefully in March and not later.

[isierra93]

That's a good news!. In the meanwhile I was using antigravity to have access to the Pro model

[sascharo]

I still don't get access to 3.1 when I authenticate with my Google account. The only way I get access right now is when I authenticate with a Gemini API key:

Authenticated with gemini-api-key /auth
 > /model
╭─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│                                                                                                                                     │
│ Select Model                                                                                                                        │
│                                                                                                                                     │
│   1. Auto (Gemini 3)                                                                                                                │
│      Let Gemini CLI decide the best model for the task: gemini-3.1-pro, gemini-3-flash                                              │
│   2. Auto (Gemini 2.5)                                                                                                              │
│      Let Gemini CLI decide the best model for the task: gemini-2.5-pro, gemini-2.5-flash                                            │
│ ● 3. Manual (gemini-3.1-pro-preview)                                                                                                │
│      Manually select a model  

[siboyapati]

I still don't get access to 3.1 when I authenticate with my Google account. The only way I get access right now is when I authenticate with a Gemini API key:

Authenticated with gemini-api-key /auth
 > /model
╭─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│                                                                                                                                     │
│ Select Model                                                                                                                        │
│                                                                                                                                     │
│   1. Auto (Gemini 3)                                                                                                                │
│      Let Gemini CLI decide the best model for the task: gemini-3.1-pro, gemini-3-flash                                              │
│   2. Auto (Gemini 2.5)                                                                                                              │
│      Let Gemini CLI decide the best model for the task: gemini-2.5-pro, gemini-2.5-flash                                            │
│ ● 3. Manual (gemini-3.1-pro-preview)                                                                                                │
│      Manually select a model  

I have the same issue and I'm on Gemini Code Assist in Google One AI Pro

[sascharo]

I have the same issue and I'm on Gemini Code Assist in Google One AI Pro

I got access now when authenticating with my Google account, but gemini-3.1-pro-preview is very busy and most of the time cannot be reached.

Command was automatically cancelled because it exceeded the timeout of 5.0 minutes without output.

[Lovexan369]

Leave replay

[aniruddhaadak80]

From my point of view, the rollout updates would be much easier for users to trust if they were paired with a short explanation of current usage limit behavior during the transition. A lot of confusion seems to come from people seeing access changes without understanding whether the limit model changed too. Even a compact note about temporary rollout constraints versus steady state behavior would help set expectations.