Gemini 3 Pro is now the new main model in Amp, powering the smart agent mode.
This is a historic change.
Ever since we've started work on Amp, back in February, Anthropic's Claude has been the main model. Not for lack of trying. We've experimented with different models for the same role: GPT-5, Gemini 2.5, Grok, and others.
But Claude has stuck, because it has been the only model that managed to strike the balance between intelligence, speed, and the willingness and ability to use tools.
Then, Gemini 3 came along. (And, on the horizon, there are even more models lined up to take the crown.)
In just a few days, the number of ecstatic messages it got in our Slack was higher than for any other model:
- "crazy, this is incredible"
- "guys this is really good, I love it, it's a ton of fun"
- "I didn't expect a model drop to affect how much I'm enjoying using amp this much"
- "Wow. It's very persistent - in a great way."
- "Hot dog!! I havenโt tried it yet but you all are getting my hopes up"
- "I'm really loving this model. Feels like a great mix of eagerness mixed with experience. Doesn't feel like a bull in a china shop."
- "am I imaginging things? it is that good, right?"
Gemini 3 checked off all the boxes that so far only Claude had checked: smart, fast, follows instructions very well, works hand-in-hand with the user if needed, very eager to use tools and uses them with high dexterity.
But, as we found out with delight, it does it all and it's better at it.
In our first internal, not-even-optimized run of Terminal-Bench 2.0 the score went up by 17 percentage points, after switching from Sonnet 4.5 to Gemini 3.
It also feels smarter, it feels better at following instructions, it's impressively clever in the way it uses tools, it cleans up after itself, it follows existing patterns in a codebase to a degree which we haven't seen before, its writing is uncannily good. It doesn't use emojis when it shouldn't. It hasn't once said that we're absolutely right. (It does have other imperfections though.)
And yet: should we really make it the default model in Amp? That's a big decision.
We went back and forth, multiple times. Every time we did, someone would invite the rest of the team: "Okay, push back, tell me I'm hallucinating, tell me why we shouldn't do it?"
But, in the end, that question was answered with Gemini 3 passing the ultimate test of all tools. When it quickly went down while early-access versions were switched, our team immediately despaired: "no, please, they can't take this away from me!"
We hope you enjoy it, too. Happy hacking!
If you want to keep using Amp with Sonnet 4.5 to help with the transition, you can do so by using a temporary configuration setting or command-line flag.
Either use the following in your Amp configuration (.config/amp/settings.json or .vscode/settings.json):
"amp.model.sonnet": true
Or use this flag when using the Amp CLI:
amp --use-sonnet
It's better, but it's not perfect.
Here is a list of issues we've seen and that we're still actively trying to prompt out of it:
- Sometimes it just "thinks" forever
- Other times, thinking-like prose leaks into the output. E.g. "I'm going to x, I did x, I'm going to y. This usually is associated with a very very large number of output tokens.
- Control characters, fake tool calls (
<todo_read></todo_read>) leaking into the output,}}at the end of the message, repeated words at the end of the message - It has sometimes been very reluctant to execute bash commands
- Unrequested git commits
- Use of non-absolute file paths
We still think that even with these issues, it's a great choice as the main model in Amp.