Google is set to launch a significant upgrade to its AI capabilities with the upcoming release of Gemini 3.8 Flash, internally codenamed “Skimaki.” According to The Wall Street Journal, the new model is expected to debut as early as this Wednesday, aiming to significantly enhance Google’s programming prowess and narrow the gap with competitors like OpenAI and Anthropic.
Internal Testing Shows Promise
In internal benchmarking tests conducted on Google’s programming tool, Jetski, engineers have shown a marked preference for Gemini 3.8 Flash over Anthropic’s Opus model. This suggests that Gemini 3.8 Flash is making substantial strides in coding performance, positioning Google more competitively in the AI development race.
A New Era for DeepMind
The rollout of Gemini 3.8 Flash coincides with a period of significant leadership changes within Google DeepMind. Last month, co-founder Demis Hassabis transitioned from daily management to focus on his roles as department chairman and Alphabet Chief Scientist. Koray Kavukcuoglu, who previously served as Hassabis’s second-in-command, has been promoted to Senior Vice President of DeepMind, taking the reins of daily operations. Kavukcuoglu has reportedly emphasized to staff his commitment to accelerating execution speed.
Addressing Past Challenges
This new model release comes after a series of setbacks for Google in the highly competitive AI model development landscape. Since the launch of Gemini 3.0 in November last year, the Gemini series has lagged behind the flagship models from Anthropic and OpenAI in critical areas such as coding. The company has also seen several prominent researchers, including Noam Shazeer, co-founder of Character.AI, and chief scientist Jeff Dean, depart over the summer.
The Gemini 3.8 Flash model belongs to the Flash series, which is designed to be smaller, more cost-effective, and faster to run, albeit with performance levels below the flagship “Pro” series models that boast trillions of parameters.
Focus on Performance and Future Development
Google’s most powerful “Pro” series models have reportedly faced delays, with a new release falling months behind schedule. An internal candidate model, 3.5 Pro, was reportedly shelved because its improvements over the Flash series were not significant enough. While the next-generation flagship model, Gemini 4, has shown positive results in pre-training evaluations, its full training is not yet complete.
To bolster the coding capabilities of its models, Google has intensified its focus on reinforcement learning over the past year, allocating more research time and computational resources to this area. Furthermore, the company recently recruited Barret Zoph, former co-founder of Thinking Machines Lab and former head of post-training at OpenAI, as Vice President of Research. Zoph will concentrate on reinforcement learning and post-training domains.
This strategic push with Gemini 3.8 Flash signals Google’s determination to regain a leading position in AI development, particularly in crucial areas like programming, by leveraging internal talent, strategic hires, and a renewed focus on accelerated development cycles.
