The artificial intelligence race is entering another competitive phase, with Google DeepMind reportedly preparing a new Gemini model aimed at improving its performance in coding and reasoning. According to a report by The Wall Street Journal, Google is developing a model that could be called Gemini 3.8 Flash, internally known as “Skimaki.”
The reported model comes as Google, OpenAI and Anthropic continue to compete for dominance in increasingly important areas such as AI-assisted software development. While Gemini 3.8 Flash is not expected to replace Google’s most powerful models, its reported coding capabilities could help Google strengthen its position among developers.
Credits: India Today
Gemini 3.8 Flash Reportedly Impresses Google Engineers
According to The Wall Street Journal, some Google engineers have reportedly preferred the upcoming model over Anthropic’s Opus model when using AI for coding tasks. If those internal results translate into strong performance on public benchmarks, Gemini 3.8 Flash could become an important addition to Google’s growing AI lineup.
Coding has become one of the most closely watched areas of AI development. Models that can understand code, identify bugs, modify software and complete complex programming tasks are increasingly being integrated into developer workflows. As a result, improvements in coding performance can have a significant impact on how useful an AI model is in the real world.
However, the reported internal preference does not necessarily mean that Gemini 3.8 Flash will outperform competing models across the board. AI systems can perform very differently depending on the benchmark, programming language, task and evaluation method.
Google Is Reportedly Increasing Its Use of Reinforcement Learning
One of the major areas of focus behind the reported model is reinforcement learning. This approach allows AI systems to improve their performance by attempting tasks repeatedly and receiving feedback about the quality of their results.
Reinforcement learning has become increasingly important as AI companies attempt to push models beyond basic instruction following. Instead of simply training models to predict the next piece of text, developers can use feedback and reward mechanisms to encourage better reasoning, problem-solving and coding behavior.
Google has reportedly been placing greater emphasis on reinforcement learning as it works to improve the performance of its Gemini models. The company has also strengthened its research team with talent experienced in post-training and reinforcement learning.
Former OpenAI Executive Barret Zoph Joins Google
Google has reportedly hired Barret Zoph, a former co-founder of Thinking Machines Lab and former OpenAI post-training leader, as a vice president of research.
Zoph has worked extensively on areas including AI training, reinforcement learning and post-training. His reported appointment highlights the growing importance of these techniques as AI companies look for ways to improve existing models after their initial training.
For Google, bringing in researchers with experience at rival AI companies could help accelerate its efforts to improve Gemini’s performance. The company already has one of the industry’s largest AI research organizations through Google DeepMind, giving it access to substantial computing resources and research infrastructure.
Coding Could Become a Major Battleground for AI Companies
The reported development of Gemini 3.8 Flash comes at a time when coding has emerged as one of the biggest commercial applications for generative AI.
AI coding tools are increasingly capable of generating functions, debugging programs, explaining unfamiliar code and handling larger software-development tasks. This has created an important market for companies developing models that can work effectively with complex codebases.
OpenAI and Anthropic have invested heavily in this area, while Google has been attempting to make Gemini more competitive for developers. A strong Flash model could therefore give Google another tool for competing in a market where speed, cost and coding accuracy are all important.
The “Flash” branding also suggests that Google may be positioning the model as a faster and more efficient option rather than simply trying to create its most powerful system. That could make it particularly useful for applications where developers need rapid responses or want to process large volumes of coding requests.
Credits: The Indian Express
Will Gemini 3.8 Flash Put Google Back in the Lead?
A successful Gemini 3.8 Flash launch would represent progress for Google, but it would not automatically mean that the company has regained the overall AI lead.
The AI model race is increasingly broad. Coding is only one measure of performance, alongside reasoning, mathematics, multimodal understanding, writing, tool use, factual accuracy and autonomous task completion. Different models can also lead in different categories.
The key test for Google’s reported model will therefore come when it is evaluated publicly. Internal testing can provide an indication of a model’s capabilities, but independent benchmarks and real-world use will offer a clearer picture of how it compares with the latest systems from OpenAI and Anthropic.
For now, Gemini 3.8 Flash remains a reported, rather than officially confirmed, development. If Google can translate the internal coding results into strong public performance, however, the model could become another important step in the company’s effort to keep pace in the rapidly evolving AI race.
