Google launches Gemini 3.8 Flash with enterprise pricing and cybersecurity variant

Google kicked off September with Gemini 3.8 Flash, its third Flash model in six weeks, alongside a cybersecurity-focused variant and expanded enterprise pricing. For builders, the key takeaway is that Google is doubling down on cheaper, specialized models for coding and agentic tasks while keeping prices flat, even as it positions Gemini as a coordination layer for multi-model workflows.

What Gemini 3.8 Flash brings to builders

Google calls Gemini 3.8 Flash its best reasoning and coding model yet, with significant improvements over 3.7 Flash in software engineering and multi-step tasks. The model is priced at 75 cents per million input tokens and $3.75 per million output tokens, matching the introductory price of the previous Flash release. That means builders get better performance without a cost increase, which matters for long-running agent loops or code generation pipelines.

The company also introduced Gemini 3.8 Flash Cyber, a variant aimed at detecting and patching software vulnerabilities at what Google claims is frontier-level performance but at a fraction of the cost of larger models. Access is initially restricted through the new Fairwind Program, which limits the model to trusted government and enterprise defenders to prevent misuse.

Enterprise pricing gets more flexible

Gemini Enterprise is adding pay-as-you-go pricing, token discounts of up to 20%, monthly caps on agent spending, and a zero-dollar base subscription option. Google is directly targeting Microsoft and Anthropic, arguing that their recurring seat fees and separate product licenses make their offerings more expensive and less flexible. For product teams, this means more predictable costs for AI features and the ability to scale agent usage without per-seat overhead.

Nearly three-quarters of Google Cloud customers already use its AI products, and those customers are spending about 50% more than their original commitments, according to Google Cloud CEO Thomas Kurian.

Why the coordination layer vision matters

Demis Hassabis, now chairman of DeepMind after a reorganization, described a future where Gemini serves as a general-purpose layer coordinating cheaper, specialized models and agents. For builders, this suggests that Google is betting on breadth and orchestration rather than just raw model quality. If Gemini becomes the router that picks the best model for each subtask, it could simplify agent architectures and reduce costs.

Caveats to watch

Analyst Gil Luria of D.A. Davidson told CNBC that while Gemini 3.8 Flash keeps Google in the race, it probably won't change the fact that Google is a distant third in the enterprise market behind Anthropic and OpenAI. Performance claims are vendor-reported, and the Fairwind program limits who can use the Cyber variant. Separately, a federal judge rejected the DOJ's bid to force Google to sell its AdX exchange, opting for behavioral remedies, which reduces regulatory overhang for Google's ad tech business but doesn't directly affect AI product decisions.