I Found Rift Runner on LMArena and I’m Pretty Sure It’s Gemini 3.0

Last night I was doom-scrolling LMArena to see if any new models dropped. In the middle of the leaderboard was a name I’d never seen before: “Rift Runner.” Naturally I clicked it.

The stats were wild. High scores across multiple benchmarks, steady ranking near the top—definitely not some random weekend project. So I kept digging, and the more I looked, the more it started to feel like Google’s next move.

First impressions

The name alone felt like a clue. Rift Runner sounds more like an internal codename than a public release. Pair that with the performance and it screamed “big company experiment,” not an indie model someone trained on the weekend.

Then there’s the timing. Google has been hinting at Gemini 3.0, but the official release is still missing. Combining that with the codename playbook Google has used before, Rift Runner started to look like a stealth drop.

Why I think it’s Gemini 3.0

A few things made the Gemini connection hard to ignore:

• The performance jump: It beats plenty of well-known models, including Gemini 2.0, across reasoning and code-heavy tasks. That kind of leap usually needs serious resources.
• The timing: We’re smack in the window where Gemini 3.0 was rumored to show up. A quiet test on LMArena fits the pattern.
• The codename vibe: “Rift Runner” just feels like one of Google’s playful internal names—think of the past “Bard” and “Chirp” codenames.

What the LMArena numbers show

I compared Rift Runner against other top models on LMArena—Claude, GPT-4.1, the usual suspects. It lands at or near the top for multimodal understanding, code generation, and long-form reasoning. The code score especially stood out versus Gemini 2.0.

Could it be something else? Sure. But all the breadcrumbs point toward Rift Runner being Google’s Gemini 3.0 test build stretching its legs in public.