In the shifting terrain of artificial intelligence, Google prepares to release a coding-optimized model called Gemini 3.8 Flash — a move that arrives not from a position of dominance, but from one of recalibration. The departure of DeepMind's cautious founder Demis Hassabis has cleared the way for a faster, more competitive posture, one that favors practical utility over monumental scale. It is a reminder that in technological races, the most consequential decisions are often not about what a tool can do, but about who decides when it is ready to be released.
Google Poised to Release Gemini 3.8 Flash Coding Model as DeepMind Leadership Shifts
Speed and cost matter as much as raw power now
Mark
So Google is releasing a coding model this week. Why does that matter right now?
Mimi
Because for most of last year, Google was the leader in AI. Then OpenAI and others moved into coding-focused models, and suddenly Google looked like it was playing catch-up. This release is Google saying they're not sitting still.
Luke
But we should be clear—this is coming from anonymous sources in the Journal. We don't have an official confirmation from Google yet. And the model is optimized for speed and cost, not raw power. That's a different competitive strategy than what OpenAI is doing.
Mark
What changed at DeepMind to make this happen now?
Mimi
Demis Hassabis, the founder, stepped down as CEO last month. He was cautious about releasing powerful models quickly. Now Sergey Brin, one of Google's founders, is more directly involved in AI strategy.
Luke
That's the narrative, anyway. But we should note that Hassabis didn't leave the company—he moved to focus on research. And Brin's involvement is also being reported by insiders, not confirmed by Google publicly.
Mark
So this is really about Google feeling pressure to move faster?
Mimi
Yes. The industry shifted toward coding AI in 2026, and Google didn't move as quickly as its competitors. This release is them trying to catch up.
Luke
And the model was tested internally on something called Jetski, and testers apparently preferred it to Claude Opus. But again, that's internal testing, not independent verification.
Mark
What does this tell us about where AI is heading?
Mimi
That speed and efficiency matter as much as raw power now. The frontier models are getting locked down by regulation. The real competition is in practical, deployable systems.
Luke
That's one reading. But we should also note that we don't know if this model will actually perform better in the real world, or if it will be adopted by developers. Internal preference doesn't always translate to market success.
The Pulse
- Google finds itself on the defensive in 2026, having ceded meaningful ground in coding AI to rivals like OpenAI and Anthropic after once leading the generative AI field.
- The departure of Nobel laureate Demis Hassabis — a voice for caution and safety — has removed a significant internal brake on the pace of product releases.
- Sergey Brin's renewed hands-on involvement signals that Google's founders are no longer content to watch competitors define the coding-AI moment.
- Skimaki, tested internally via a tool called Jetski, reportedly outperformed Anthropic's Claude Opus in developer preference trials — suggesting real competitive teeth beneath the modest branding.
- Rather than chasing raw parameter scale, Google is betting on speed and cost-efficiency, positioning Gemini 3.8 Flash as a practical workhorse for the token-economy era.
In the shifting terrain of artificial intelligence, Google prepares to release a coding-optimized model called Gemini 3.8 Flash — a move that arrives not from a position of dominance, but from one of recalibration. The departure of DeepMind's cautious founder Demis Hassabis has cleared the way for a faster, more competitive posture, one that favors practical utility over monumental scale. It is a reminder that in technological races, the most consequential decisions are often not about what a tool can do, but about who decides when it is ready to be released.
Google is preparing to release Gemini 3.8 Flash — internally called Skimaki — as early as this week, according to sources familiar with the matter. The release comes at a pivotal moment: DeepMind founder Demis Hassabis stepped down as CEO last month, shifting the organization's center of gravity toward faster, more product-driven decision-making.
For much of the recent past, Google held a strong position in generative AI. But the rise of agentic coding platforms in 2026, led by tools that made token-intensive coding assistance a core feature, reshuffled the competitive order. OpenAI moved aggressively into the space in March, and Google has been under pressure to respond.
Skimaki is not designed to be a frontier model in the traditional sense — it carries no record-breaking parameter count and requires none of the regulatory scrutiny now attached to the largest AI systems. Instead, it is built for speed and cost-efficiency, qualities increasingly prized as aggregation platforms like OpenRouter make practical performance and expense central concerns for developers. Internal testing via a Google tool called Jetski found that developers preferred Skimaki over Anthropic's Claude Opus, a result that suggests genuine real-world advantage.
Hassabis, a Nobel Prize winner in chemistry, had long urged restraint in releasing powerful AI, citing safety concerns. His new role focuses on scientific research, leaving product strategy to others. With Sergey Brin reportedly pushing for faster Gemini releases, the organizational message is clear: Google is done waiting. Whether Gemini 3.8 Flash proves enough to reclaim competitive standing in the coding-AI race remains an open question — but the willingness to move is itself a signal.
Google is preparing to release a new coding-focused artificial intelligence model called Gemini 3.8 Flash, internally known as Skimaki, as early as this week, according to anonymous sources who spoke with the Wall Street Journal. The timing coincides with a significant leadership shift at Google DeepMind, where founder Demis Hassabis stepped down as CEO last month.
The move marks a strategic pivot for Google in response to what many in the industry perceive as the company losing ground in the AI race. For much of last year, Google held a commanding position in generative AI. But the emergence of agentic coding platforms—particularly OpenClaw, which gained traction in 2026 by making token-intensive coding assistance a central feature—shifted the competitive landscape. OpenAI moved into this space in March, and Google has faced mounting pressure to follow suit.
Skimaki is not designed to be a frontier model in the traditional sense. It will not carry the massive parameter counts of unreleased systems like OpenAI's Astra, nor will it require the kind of White House vetting that now accompanies the largest AI systems. Instead, the model prioritizes speed and cost-efficiency—attributes that have become increasingly valuable as the industry develops platforms like OpenRouter, which aggregate tokens from multiple models to optimize performance and expense. This represents a deliberate choice to compete on practical utility rather than raw scale.
The internal testing has been conducted using a Google coding tool called Jetski. According to people familiar with the results, developers who tested Skimaki preferred it to Anthropic's Claude Opus, suggesting the model may have genuine competitive advantages in real-world coding tasks.
The leadership change at DeepMind appears to be the catalyst for this acceleration. Hassabis, a Nobel Prize winner in chemistry, had publicly expressed caution about releasing revolutionary AI models, citing safety and risk concerns as recently as last year. His new role focuses on scientific research rather than product development. Meanwhile, Google founder Sergey Brin has taken a more active interest in AI strategy, with reports indicating he is pushing for faster releases of competitive Gemini models. Insiders suggested Hassabis had been gradually stepping back for roughly a year, but the formal transition signals a clear change in organizational priorities.
The release of Gemini 3.8 Flash, if it arrives as expected, would represent Google's answer to the coding-AI moment that has defined 2026. Whether the model's speed and efficiency advantages prove sufficient to restore Google's competitive standing remains to be seen, but the move suggests the company is no longer willing to wait.
Notable Quotes
Google insiders told the Journal that Skimaki has been tested on an internal coding tool and that developers prefer it to Anthropic's Claude Opus— Anonymous sources to Wall Street Journal
Want the full story?
Read the original at Gizmodo ↗
Technology