Google has introduced two new artificial intelligence models under its Gemini 3.8 series, rolling out Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. The technology firm describes the pair as its most capable releases to date for coding and logical reasoning, combining rapid processing with reduced operating costs. Both models share an underlying intelligent architecture refined with cybersecurity training, relying on autonomous systems that iteratively assess their outputs to handle complex, multi-stage assignments.
Engineering Upgrades and Unchanged Pricing
Gemini 3.8 Flash is designed primarily for multi-step reasoning, autonomous agent workflows, and software development. Google confirmed that standard pricing remains unchanged from the previous edition, fixed at $0.75 per million input tokens and $3.75 per million output tokens. According to Google Chief Executive Sundar Pichai, the model delivers measurable progress over the earlier 3.7 Flash across software engineering and agentic functions. The company stated that Gemini 3.8 Flash surpassed several larger frontier systems on the DeepSWE v1.1 evaluation, a test devised to track how effectively artificial intelligence manages lengthy programming tasks autonomously from beginning to end.
Specialised Security Model and Fairwind Access
The second release, Gemini 3.8 Flash Cyber, is engineered exclusively for digital defence and defensive security workflows. Available to verified professionals and institutions through Google’s new Fairwind programme, the software focuses on vulnerability identification and remediation rather than offensive cyber operations. Pichai reported that the Cyber edition scored 86.2 percent on the CyberGym industry benchmark and 47.2 percent on CWE-Bench for vulnerability patching. In internal evaluations spanning 20 programming languages, Google claimed the system achieved an identification success rate exceeding 70 percent, compared against reported benchmark figures of 83.8 percent for Anthropic’s Opus 5 and 83.6 percent for GPT-5.6 Sol under differing testing conditions.
Real-World Application in Browser Defence
Beyond theoretical testing, Google stated that the cybersecurity edition is actively applied in live environments. Assessments carried out by the Chrome security team indicated that the model generated 2.6 times more accurate fixes for actual browser flaws than larger commercial alternatives. Pichai noted that these practical defensive results justified granting early deployment access to government agencies and commercial partners under the Fairwind initiative. The dual launch signals Google’s intent to pair cost efficiency and high processing speed with specialised tools for software creation and autonomous system security.
