Google launches Gemini 3.8 Flash with cyber variant
Google unveiled Gemini 3.8 Flash, its third Flash model in six weeks, available in standard and Cyber-tuned versions. The model shows improved reasoning and coding performance, topping the DeepSWE leaderboard at a discounted introductory price. Google continues to prioritize Flash releases while pausing updates to its Pro line.
Google’s latest Flash release marks the third such model in six weeks, signaling a strategic shift away from its larger Pro line, which has seen no updates since early 2026. The new model is offered in two forms: a general-purpose “workhorse” and a Cyber variant tuned for security tasks. Pricing is temporarily discounted, with input tokens at $0.75 per million and output at $3.75, before rising to $1.50 and $7.50 respectively. The Cyber edition is restricted to trusted testers and government partners, while the standard version rolls out across Google’s ecosystem today, requiring a paid subscription for app access but free via AI Studio.
The model’s benchmark results show modest gains over its predecessor in most areas, with larger improvements in coding. It now leads the DeepSWE leaderboard for software engineering problem-solving, and Google reports that internal security teams saw a 2.6x increase in patch accuracy when using the Cyber variant. The company also cites partner testimonials from Wiz and Palo Alto Networks, though independent verification remains pending.
The rapid cadence of Flash releases could reshape developer expectations around AI model pricing and capability, as smaller, cheaper models increasingly rival frontier systems in specialized tasks like coding. Businesses may benefit from lower token costs, but the pause on Pro updates might signal a plateau in large-model innovation, potentially slowing enterprise adoption of AI for complex reasoning. Cybersecurity teams could gain powerful tools, yet reliance on AI for vulnerability patching introduces new risks if models err, affecting trust in automated defenses.