According to media reports, the release of Google's flagship and most powerful model, Gemini 3.5 Pro, has been delayed by several months from the original plan. The primary reason is that Google is attempting to enhance the model's programming capabilities, but progress has fallen short of expectations. Recently released new models from OpenAI and Meta have further surpassed Google's existing products in AI code generation capabilities. Following the news, Google's stock price dropped as much as 5% intraday, ultimately closing down 4.4%.
Google is facing increasingly severe internal pressure in the AI race. The release of its most powerful flagship model, Gemini 3.5 Pro, has been delayed by several months from the original schedule. Meanwhile, competitors have successively surpassed it in programming capabilities, making Google's market position increasingly passive.
According to Bloomberg, citing informed sources, the main reason for the delay is that Google is trying to improve the model's programming abilities, but progress has been disappointing. At the end of last month, Google updated its training data to address this weakness; however, the results were unsatisfactory.
New models recently released by OpenAI and Meta have further exceeded the AI code generation capabilities of Google's current offerings.
This situation has sparked widespread dissatisfaction within Google. According to ten current and former employees, many engineers, AI researchers, and management personnel are concerned that Google is losing its market edge. The frustration among some researchers has evolved into a wave of departures, with top labs like Anthropic becoming the primary destinations.
After the news broke, Google's stock price plummeted intraday, falling as much as 5%, and ultimately closed down 4.43%.
Organizational Complexity Slows Release Pace
Google's delays in releasing AI models are closely linked to the internal coordination challenges posed by its vast product portfolio. According to informed sources, Google's model release process involves multiple layers of stakeholders and requires integrating AI capabilities into numerous product lines such as Search, Maps, and YouTube, which inherently consumes time.
A former employee likened this coordination dilemma to "boiling an ocean." When departments have misaligned directions and tasks are redundantly pushed by multiple teams, maintaining a unified product strategy becomes increasingly difficult, and no single product can secure sufficient resources to achieve a market breakthrough.
A Google spokesperson responded, stating that the company is "moving quickly on multiple model families while maintaining a high value-to-cost ratio for customers." They added that Google is currently testing 3.5 Pro, an upgraded Flash model, and other models with partners, while actively engaging with the U.S. government on model testing and a broader industry standards framework.
Programming Capability Competition Reflects Internal Divisions
AI code generation capability has become a core battleground in the current model competition, yet Google's efforts in this area are hampered by internal constraints. According to former employees, Google co-founder Sergey Brin and others actively pushed for the company to accelerate its AI programming initiatives, but related efforts were hindered by internal factional struggles.
Currently, Google Cloud, Google DeepMind, and the Android team are independently developing AI programming tools for developers, with some consumer product teams also involved. This multi-pronged approach further disperses resources.
Internal resistance is also significant. Some engineers hold the principled stance that "code should be written by humans" and are skeptical of AI code generation. During the early promotion phase, due to security concerns about proprietary code leaking into training data, Google once restricted employees from using Gemini to write or analyze code. Although this policy has been relaxed, it missed numerous internal experimentation opportunities.
Google stated that at the recent Cloud conference, the company announced that 75% of its internal code is currently AI-generated and, after review, has successfully entered production environments, meeting Google's standards.
To reduce internal coordination chaos, Google's chief AI architect, Koray Kavukcuoglu, is collaborating with the main engineering team to promote the integration of internal AI programming tools. DeepMind also formed a new team earlier this year dedicated to AI programming, led by research engineer Sebastian Borgeaud.
Current Gemini 3.5 Flash Receives Mixed Reviews
Prior to the official release of 3.5 Pro, Google's existing Gemini 3.5 Flash has left a mixed reputation in the market. Rodrigo Davies, a product manager at design platform Figma, stated that Figma has integrated 3.5 Flash into its newly launched AI assistant "Figma agent," believing the model strikes a good balance between speed and quality.
However, Freddy Vega, founder and CEO of the Latin American online education platform Platzi, offered a starkly different assessment. He pointed out that 3.5 Flash is in an awkward position: it is more expensive than the previous generation 3.1 Flash, yet slower, and still lags significantly in capability compared to competitors' flagship products. It also performs poorly in handling structured data.
He noted that his team has shifted from Google to Anthropic's mid-tier model, Claude 3.5 Sonnet, for tasks requiring both speed and reasoning capabilities.
AI researchers generally believe that Gemini's current biggest competitive advantage lies in its ability to leverage Google Search data. However, in the dimension of building the most powerful model, Anthropic and OpenAI have taken the lead. Google stated that the company still holds differentiated advantages in areas like multimodal input processing and AI world models, but whether these advantages can be translated into market share remains to be seen.
Comments