Gemini 3.8 Flash is now available, marking Google’s third Flash model release in just six weeks. The rapid cadence stands in contrast to Google’s frontier tier, which has not seen a new Gemini Pro model since early 2026. That gap raises the possibility that the previously promised Gemini 3.5 Pro may never ship. Google, meanwhile, describes 3.8 Flash as its strongest model yet for reasoning and coding tasks.
The release arrives in two versions. The standard Flash model is positioned as a general-purpose “workhorse” suited to everything from agentic workflows to software development. A second variant, Gemini 3.8 Flash Cyber, is built on the same foundation but tuned specifically for vulnerability detection and mitigation. The Cyber model replaces the earlier 3.5 version.
Pricing and Benchmark Performance
Developers get the same offer Google extended for the 3.7 Flash launch a couple of weeks earlier. API access is available at an introductory rate through the end of the year: EUR 1 per million input tokens and EUR 3 per million output tokens. Standard pricing will rise to EUR 1 and EUR 6 respectively, though new models are likely to appear before that change takes effect. The discounted pricing reflects broader industry pressure, as competing AI labs have recently cut token costs to retain cautious business customers.
Google’s published benchmarks position Gemini 3.8 Flash as competitive with, and in some cases ahead of, larger and more expensive models. The results show marginal gains over Gemini 3.7 Flash across most tests, with more significant improvements in coding evaluations. The model now leads the DeepSWE leaderboard, which measures the ability to solve complex software engineering problems, and does so at a lower cost under the current discounted rate. Google reportedly held back Gemini 3.5 Pro when its coding results fell short of rivals, but these new Flash figures suggest the smaller models are now competing with market leaders.
Computer use remains a weak spot. While 3.8 Flash improves on 3.7 Flash in the OSWorld-2.0 test of agentic computer use, it still trails Claude Opus by a wide margin. GPT models also perform poorly on that particular benchmark.
Cybersecurity Gains and Availability
Gemini 3.8 Flash Cyber targets a specialized field, and Google reports substantial improvements in internal testing. The model identified more vulnerabilities and produced working patches more frequently than its predecessors. The Chrome security team recorded a 2.6x increase in patch accuracy, while the Cloud team reports that the model located a critical vulnerability in just two hours. Google also cites statements from partners including Wiz and Palo Alto Networks supporting the model’s capabilities.
Availability differs by version. Gemini 3.8 Flash rolls out across the Google ecosystem starting today, while Gemini 3.8 Flash Cyber is currently restricted to trusted testers and government users. As with the prior Flash release, access requires a Pro or Ultra subscription.
Source
Image: arstechnica.com
