"Performance and its stunning," Google’s large model rarely creates "momentum" before its release, will Gemini 3.0 debut this week?

Google's upcoming Gemini 3.0 artificial intelligence model is attracting widespread attention in the industry.

Prediction markets indicate that the model will be launched next week, and CEO Sundar Pichai responded to related speculation on social media with a "thinking emoji," almost confirming this timeline. This marks the first time Google has conducted such a large-scale internal and external promotional campaign before the release of a large model.

Moreover, those who have interacted with the model have given it high praise. According to Business Insider on Monday, insiders described the new model as "extremely impressive," expecting significant improvements in coding and multimedia content generation. Google employees have begun to express their excitement about the release on social media, a phenomenon that has been rare before previous model launches.

The test results of the model in professional fields show groundbreaking progress. Mark Humphries, a history professor at Laurentian University in Canada, tested a suspected unreleased version of Gemini 3.0 through Google AI Studio and found that it was nearly perfect in recognizing 18th-century handwritten manuscripts, with a character error rate of only 0.56% and a word error rate of 1.22%, representing a 50%-70% improvement over the previous generation Gemini 2.5 Pro, reaching expert-level human performance.

For Google, which had been on "red alert" since the release of ChatGPT at the end of 2022, Gemini 3.0 is seen as a key step in reshaping its market position, especially in the context of OpenAI's highly anticipated ChatGPT-5 release, which failed to produce significant immediate impact.

Rare Pre-Release Hype

The atmosphere surrounding Google's upcoming large model release is noticeably different from previous occasions. Prediction markets have begun to bet that Gemini 3.0 will be released next week, and Pichai responded to discussions on social platform X with a "thinking emoji," a subtle gesture widely interpreted as confirmation of the release timeline.

Google employees' active presence on social media is also unusual. Many employees have openly expressed their excitement about the new model's release, a collective pre-release behavior that has not been common in previous model launches. Not only insiders but also many external individuals who have learned about the model's capabilities have shared enthusiastic reviews online.

According to Business Insider, insiders have described the model as "extremely impressive," expecting significant enhancements in coding and multimedia content generation, potentially including a major upgrade to Google's popular image tool NanoBanana.

Professional Testing Shows Breakthrough Capabilities

Mark Humphries' testing provides concrete examples for understanding the new model's capabilities. He used his professional work—analyzing 18th-century handwritten accounting ledgers—as a benchmark test. This task is extremely challenging, requiring not only the recognition of messy handwriting but also the integration of historical context, linguistic nuances, and logical reasoning.

Humphries pointed out that interpreting historical handwritten texts requires abilities that go beyond visual recognition. "When you go back in time, you enter a different realm. People spoke differently, used unfamiliar vocabulary, or used familiar words in unfamiliar ways." In the past, people used different measurement and accounting systems, different wording, punctuation, capitalization, and spelling."

Test results show that the character error rate of the previous generation Gemini 2.5 Pro on these complex documents was about 4%, roughly equivalent to the level of professional human transcribers. The new model reduces the character error rate to 0.56% and the word error rate to 1.22%, achieving expert-level human performance standards.

More notably, the model demonstrates reasoning abilities. Humphries found that the model could spontaneously perform step-by-step symbolic reasoning, such as inferring "145" as "14 pounds 5 ounces" in an 18th-century merchant's ledger, which is not just text recognition but an understanding of the economic and cultural systems that generated these records.

A Turning Point in Google's AI Strategy

For Google, the release of Gemini 3.0 is strategically significant. Since the release of ChatGPT at the end of 2022, Google has been seen as playing catch-up in the AI race, even issuing an internal "red alert." Business Insider cited insiders stating that the new model may give Google a chance to regain a leading position, especially after OpenAI's ChatGPT-5 failed to make an immediate significant impact.

The model is expected to achieve significant improvements in coding and multimedia content generation. Google's image generation model NanoBanana has recently received positive feedback from users, with the name originating from a placeholder created by an employee named Nina. According to David Sharon, product manager for Google Gemini App, the name was used when Google anonymously submitted the model to the open AI evaluation platform LM Arena for fair testing, resulting in unexpected popularity in online communities, leading Google to officially adopt the name.

The most profound implication is that if the capabilities of the new model are systematically validated, AI may be transitioning from complex "random parrots" to systems with genuine understanding capabilities. Humphries noted, "If this behavior proves to be reliable and replicable, it points to something profound: true reasoning may not require explicit rules or symbolic frameworks to emerge, but can instead emerge from scale, multimodality, and exposure to sufficient structured complexity."

For historians, near-perfect handwritten text recognition combined with contextual understanding will allow for the rapid digitization and analysis of knowledge trapped for centuries, potentially rewriting our understanding of the past. For broader applications, reasoning-capable AI could begin to automate complex cognitive tasks previously thought to be the exclusive domain of human experts