Google released Gemini 3.8 Flash on September 2, its third Flash-tier model in six weeks, alongside a locked-down cybersecurity variant built for patching software vulnerabilities. The company positioned 3.8 Flash as its new workhorse model for coding, AI agents, and multi-step reasoning, according to Google’s official announcement.

What’s new

Google said 3.8 Flash outperforms larger frontier models on DeepSWE v1.1, a benchmark for long-horizon software engineering tasks, and beats its predecessor, Gemini 3.7 Flash, on the Vals Finance Agent V2 and Harvey Legal Agent benchmarks. On HLE-Verified, a test of multi-step reasoning, the model scored 54.9%. Google attributed the gains to the model “working harder” on complex problems, running more reasoning steps and calling tools iteratively before answering, with adjustable thinking levels that let developers trade speed for accuracy.

A cybersecurity sibling

Alongside the general-purpose model, Google shipped Gemini 3.8 Flash Cyber, a variant restricted to the company’s Fairwind Program for governments, critical-infrastructure operators and open-source maintainers. Google’s Chrome security team said the Cyber model produced 2.6 times more correct patches for Chrome vulnerabilities than the best commercial models available, despite those rivals being much larger.

Pricing and availability

3.8 Flash is live for developers through Google Antigravity, AI Studio and Android Studio, for enterprises through Gemini Enterprise, and for consumers through the Gemini app, AI Mode in Search and Gemini in Sheets for Google AI Pro and Ultra subscribers. Google kept the same introductory price it used for 3.7 Flash — $0.75 per million input tokens and $3.75 per million output tokens — through the end of 2026, rising to $1.50 and $7.50 respectively on January 1, 2027.

The rapid cadence, a third Flash release in six weeks, underscores how aggressively Google, OpenAI and Anthropic are iterating on mid-tier models built for coding and agentic workflows rather than chasing headline benchmark scores alone.