Mysterious New Google Models Appear on LMArena, Hinting at Gemini 3.0 Pro?

Two mysterious Google models named ’lithiumflow’ and ‘orionmist’ recently appeared on the AI arena LMArena. All signs point to this being the highly anticipated Gemini 3.0 Pro, whose powerful performance and special abilities have sparked heated discussions in the community.

Recently, on the well-known AI model competition platform LMArena, two new faces from Google quietly appeared: ’lithiumflow’ and ‘orionmist’. This discovery immediately caused a stir in the AI enthusiast and developer community. Everyone is guessing, could this be the long-rumored next-generation flagship model from Google—Gemini 3.0?

All signs seem to point to this answer.

Code Names Leaking Secrets? The Possibility of Gemini 3.0

Insiders familiar with Google’s naming conventions quickly noticed the clues. It is rumored that the codename ‘orion’ has always been associated with the development of Gemini 3 within Google. The appearance of the ‘orionmist’ model this time naturally leads people to associate it with the Gemini 3 family.

Furthermore, according to some online discussions and analysis, it is widely speculated that ’lithiumflow’ may be the Gemini 3.0 Pro version, while ‘orionmist’ corresponds to the lighter Flash version. Although Google has not officially confirmed this, this method of testing in disguise is a common practice in the AI industry.

Not Just Strong in Benchmarks, Special Skills are Maxed Out

The quality of a model depends on its capabilities. Judging from the feedback of some lucky users who have encountered the new models on LMArena, the performance of ’lithiumflow’ and ‘orionmist’ has not disappointed.

In some preliminary benchmark tests, such as simplebench, the new models scored as high as 8-10 points (out of 10), significantly surpassing the existing Gemini 2.5 Pro. This means a significant leap in comprehensive abilities such as logical reasoning, code generation, and common sense Q&A.

However, what users are most excited about are some of its ‘special talents’:

• Superb Role-Playing: This is definitely good news for users who enjoy role-playing interactions with AI. The new model’s role-playing ability far exceeds that of its predecessors. It can accurately grasp the tone, personality, and background settings, providing an immersive experience.
• Powerful SVG Processing Capabilities: Another amazing highlight is its ability to process Scalable Vector Graphics (SVG). You can ask it to generate an SVG image of a ‘pelican riding a bicycle’, and it can not only understand this slightly absurd command but also produce well-structured and stylish SVG code. This is a task that has often been a headache for many top models in the past.
• HTML Content Generation: In addition to SVG, the new model can also handle HTML content, such as generating a weather card or a simple webpage model of a catapult. This demonstrates its potential in front-end code generation and multimodal understanding.

It is worth mentioning that even with the significant functional enhancements, the new model’s Context Length remains at an astonishing 1 million tokens. This means it can process and remember an extremely large amount of information, which is crucial for complex tasks such as analyzing long reports and codebases.

How to Experience It Yourself?

Want to try these two mysterious models for yourself? The method is simple. Go to the LMArena.ai website and click on the ‘Battle’ mode. The system will randomly assign two anonymous models for you to compare. If you are lucky enough, you may encounter the new Google models codenamed ’lithiumflow’ or ‘orionmist’.

Frequently Asked Questions (FAQ)

Q: Are these two new models confirmed to be Gemini 3.0 Pro?

A: Currently, Google has not made an official announcement. However, based on its internal codename ‘orion’ and its strong performance on LMArena, the community widely believes that this is very likely Gemini 3.0 Pro or its early test version.

Q: How is the new model stronger than Gemini 2.5 Pro?

A: From preliminary tests, the new model surpasses Gemini 2.5 Pro in overall performance, especially showing significant advantages in specific areas such as role-playing, SVG image generation, and HTML content processing.

Q: How can I use these new models?

A: Currently, the only way to experience these models is through the ‘Battle’ mode on the LMArena platform, where you can encounter them through random pairing. It is expected that Google will officially integrate them into Gemini Advanced or other product lines in the near future.