.png)
2 weeks ago
1
Mysterious, high-performing AI models 'lithiumflow' and 'orionmist' have appeared on LMArena, fueling speculation they are pre-release versions of Google's Gemini 3.0.
Google is fueling speculation that its Gemini 3.0 AI is on the verge of release after two mysterious new models appeared on the public testing platform LMArena.
Spotted over the weekend, the models, codenamed ‘lithiumflow’ and ‘orionmist’, are widely believed to be pre-release versions of Gemini 3.0 Pro and Flash.
The timing aligns with a rumored October 22 announcement, a date that first surfaced via an alledged leaked internal timeline document.
Early testers report the models show a remarkable leap in capabilities, particularly in generating complex graphics from code and advanced visual reasoning. This suggests Google is in the final stages of preparing its next major challenge to rivals.
New Mystery Models Appear in the Arena
Over the weekend, users on platforms like X and Reddit began sharing screenshots of ‘lithiumflow’ and ‘orionmist’ in live matchups.
LMArena, run by LMSYS, is a well-known platform where new models are often anonymously tested before an official launch, a process that helps benchmark them in real-world scenarios.
While the models don’t appear on the public leaderboard yet, their presence in live battles is confirmed by multiple independent sources.
Community analysis quickly link the codenames to Google. The “Orion” theme has been associated internally with Gemini 3, and the evocative names fit Google’s style.
The prevailing theory maps ‘lithiumflow’ to the more powerful Gemini 3.0 Pro and ‘orionmist’ to the faster, more efficient Gemini 3.0 Flash.
X users that have been A/B testing Gemini 3.0 checkpoints via AI Studio or the Gemini app, such as cannn064 and Lentils80 suggest the models differ by ‘orionmist’ having web search (grounding) enabled, while ‘lithiumflow’ does not.
🚨 Gemini 3.0 Pro is finally on LM Arena!
Orionmist → Gemini 3.0 Pro with Grounding (web search) ON
Lithiumflow → Gemini 3.0 Pro with Grounding OFF
Are they nerfed? Another Zenith situation? 👀
We’ll find out soon.
Here are some outputs below ⬇️ pic.twitter.com/hgVKX8fA58
— can (@cannn064) October 19, 2025
What Can ‘Lithiumflow’ and ‘Orionmist’ Do?
The most striking feedback from early testers centers on the models’ advanced generative and visual skills. Multiple users have highlighted their exceptional ability, particularly in generating complex graphics-as-code, such as high-quality Scalable Vector Graphics (SVGs).
⛱️Testing the new Gemini 3 checkpoints
Here are some pelicans it generated🦩 pic.twitter.com/hxIFIgEC6T
— can (@cannn064) October 19, 2025
One popular stress test, prompting the AI to create an SVG of “a pelican riding a bicycle,” yielded remarkably clean and coherent results from ‘lithiumflow’. Another user showcased its impressive output with a detailed PS4 controller graphic. These capabilities represent a significant step up from many current-generation models.
lithiumflow (Supposedly Gemini 3 Pro on LMArena) can perfectly tell the time when given an image of a clock
May be SOTA for vision? Will be doing some heavy tests with this model
c @synthwavedd pic.twitter.com/liYD8H24QE
— bradley (@Neuralithic) October 19, 2025
Beyond code, the models demonstrate superior visual understanding. A report highlighted ‘lithiumflow’ accurately reading the time from an image of an analog clock—a task where many AIs still falter.
Some testers noted weaker-than-expected coding performance, but this could be due to the models being ‘nerfed’ or throttled for public testing. Other testers say coding performance is better than with other state of the art models such as Claude Sonnet 4.5.
🚨Gemini 3 – [LithiumFlow] Echoes: Luminous
I was blown away by this. It uses mathematics to reduce code size and employs many algorithms and equations. It deeply understands concepts and executes them in one shot. I created this game based on my other game.
CodePen next pic.twitter.com/eDP84QmayB
— ص (@sbalhatlani) October 19, 2025
Connecting the Dots to a Gemini 3.0 Launch
The LMArena sightings have reignited chatter around a rumored October 22 announcement date. Google has not officially confirmed any date.
Adding fuel to the fire is a cryptic social media post from Logan Kilpatrick, Google’s product lead for AI Studio. On October 19, he tweeted three ship emojis without comment. However, this came just one day after a major AI Studio overhaul, which his cryptic post is possibly referring to.
🛳️🛳️🛳️
— Logan Kilpatrick (@OfficialLoganK) October 19, 2025
Addind to the speculation, the official post from Google about the AI Studio update concluded with a powerful teaser, stating, “this week was about building a better foundation. Next week, we’re going to build something magical on top of it.”
This sequence of events strongly implies that significant product launches are imminent. While not a direct confirmation, it aligns perfectly with the rumored timeline for a Gemini 3.0 reveal.
The Broader AI Race
This flurry of activity reflects the intense pressure Google faces in the competitive AI market. The company is in a clear arms race with rivals who have recently unveiled powerful new models. In September, OpenAI launched GPT-5-Codex, a specialized model for “agentic coding”.
Shortly after, Anthropic released Claude Sonnet 4.5, which it claimed was the world’s best coding model. Michael Truell, CEO of Cursor, praised its capabilities, stating, “we’re seeing state-of-the-art coding performance from Claude Sonnet 4.5, with significant improvements on longer horizon tasks.”
Google’s Gemini 3.0 must not only match but exceed these benchmarks to reclaim a leadership position.
The new models are also central to Google’s enterprise strategy. Gemini 3.0 Pro is expected to power Gemini Enterprise, a platform for businesses to build custom AI agents.
Google Cloud CEO Thomas Kurian has framed this as a key differentiator, arguing that unlike rivals, “some companies offer AI models and toolkits, but they are handing you the pieces, not the platform. They leave your teams to stitch everything together.”
This vision was echoed by Google CEO Sundar Pichai, who stated, “Gemini Enterprise is designed on the premise that true business transformation in the era of AI must go beyond simple chatbots.”
For official context, Pichai recently stated the next Gemini version would arrive ‘later this year’. Gartner analyst Chirag Dekate noted that “how Google is able to leverage this unified messaging in the Gemini 3.0 launch sequence… will also be a crucial litmus test.” For now, the industry watches to see if this quiet rollout will soon become a roar.
Markus has been covering the tech industry for more than 15 years. He is holding a Master´s degree in International Economics and is the founder and managing editor of Winbuzzer.com.
