Google发布Gemini 3.8对AI市场意味着什么

Generative AI Foundation Models News What Google’s Release of Gemini 3.8 Says About the AI Market With the new model, Google focused on long-horizon reasoning tasks, while also boosting cyber performance. It also introduced a cyber program similar to that of other AI vendors. Esther Shittu,News Writer, AI Business September 3, 2026 3 Min Read winhorse via Getty Images While Google’s release of Gemini 3.8 shows how the model maker is strengthening its coding, agentic and cyber capabilities, it also highlights the lack of differentiation in the AI market and the trend toward exclusive cyber programs. The search and AI giant introduced Gemini 3.8 on Wednesday, framing it as its best reasoning and coding model yet. Gemini 3.8 has two variants: Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. The 3.8 Flash model is built for long-horizon software engineering, autonomous agents and complex enterprise workflows. The 3.8 Flash Cyber version is designed for autonomous vulnerability discovery and automated code patching. Google said the model is accessible to a trusted group of users using the new Fairwind Program. The initiative provides Google’s most powerful AI and cyber defense capabilities to trusted customers, government agencies and cybersecurity partners. Google’s Fairwind Program is similar to Anthropic’s Project Glasswing, which made Anthropic’s most powerful model, Mythos, available to a trusted group of users, and to OpenAI’s Project Daybreak -- a program to assist selected cyber defenders. The move to create a private program to keep powerful AI cyber models out of the hands of hackers or others who could misuse them is another way Google is competing with Anthropic and OpenAI. Still, it shows that all the model makers are moving in tandem. Related:OpenAI Touts GPT-6 Astra as Its Safest Model, But It's Still Dangerous “Everybody is gravitating toward the same things,” said Arun Chandrasekaran, an analyst at Gartner. “Like coding, everybody is trying to over-index the model on coding. In cybersecurity, now everybody is trying to really optimize these models for cybersecurity use cases.” Lack of Differentiation He added that, with frontier model makers adhering to similar strategies, little about them is distinctive. “There’s differentiation in terms of capabilities, but in terms of how they convey what these models are good at, what use cases, what problems they are solving, all of these companies seem to be doing very similar things,” Chandrasekaran added. He continued that true differentiation comes from going deeper into specific business functions, for example, such as supply chain or life sciences. “Those are opportunities for differentiation because you’re not building a general-purpose system,” he said. “You’re building something very deep, that’s very specific to an industry.” Pricing War Continues Nevertheless, for Google specifically, the release of Gemini 3.8 allows it to respond to its competitors' moves in both pricing and capabilities. As with the 3.7 iteration, Google is offering an introductory price for 3.8 of $0.75 per million input tokens and $3.75 per million output tokens for Gemini 3.8 Flash. By comparison, OpenAI GPT-5.6 Luna costs $1 per million input tokens and $6 per million output tokens. Related:Prompt: Nvidia Moves up the AI Stack “It remains to be seen whether they can sustain that kind of pricing,” said Sid Nag, founder and analyst at Tekonyx. “All this is really all about pushing down the cost curve. Everyone is focused on the cost for a million input tokens.” He added that with Gemini 3.8 Flash, Google shows that “token pricing is no longer sufficient to measure inference economics.” This dynamic is evident in Google's advice that customers should still choose Gemini 3.7 Flash for workloads that require greater efficiency, meaning that, despite the token cost, enterprises might sometimes save more tokens by using a smaller model. “Google is admitting that this particular release 3.8 Flash consumes more tokens at higher reasoning levels,” he continued. By offering Gemini 3.8 Flash Cyber only to trusted users, Google is following its rivals in trying to create a “moat” around its AI model technology, Nag said. However, the exclusivity approach with cyber models is creating tension in the AI industry, as it can lead to a lack of openness and transparency. While open models have been criticized for their ease of manipulation by hackers, they have still spurred significant innovation. With Google’s release and other exclusive cyber programs, the AI industry continues to navigate the tension between security and openness. Related:Enterprise AI Startup Wonderful Now Valued at $5B “It remains to be seen where the industry will skate to,” Nag said. About the Author Esther Shittu News Writer, AI Business Esther Shittu has covered AI technologies and industry trends since 2021. As co-host of the Targeting AI podcast, she talks with experts, thought leaders and practitioners exploring critical AI developments. Before AI Business, she wrote for SearchEnterpriseAI, the New York Daily News, Bklyner and the Brooklyn Daily Eagle. When she's not diving deep into the world of AI, she spends her time on passion projects and raising her three daughters. See more from Esther Shittu Want more AI Business stories in your Google search results? Add Us Now Sign Up for the Newsletter The most up-to-date AI news and insights delivered right to your inbox! Stay Updated!