Google launches Gemini 3.8 Flash and 3.8 Flash Cyber: third Flash release in six weeks targets reasoning, coding and cybersecurity

First, positioning and pricing. Gemini 3.8 Flash continues the Flash line's 'low price, high speed' approach: the official blog states it delivers 'our best reasoning and coding model yet at the same speed and low cost of 3.7,' with performance that often approaches far more expensive frontier models. On price, 3.8 Flash keeps 3.7 Flash's introductory rate — $0.75 per million input tokens and $3.75 per million output tokens — but that promo pricing expires on December 31, 2026, adjusting to $1.50 input and $7.50 output per million tokens from January 1, 2027. The cadence is a clear signal: Google is using 'more capability, same price' to pressure rivals like OpenAI on the volume-oriented Flash line. Google also notes that developers prioritizing compute efficiency can use lower effort levels to minimize token overhead, or keep relying on the fully supported 3.7 Flash.

On general capability, 3.8 Flash delivers an impressive report card. Official data shows that on DeepSWE v1.1, a benchmark measuring long-horizon software engineering, 3.8 Flash autonomously solves complex engineering problems end to end at a fraction of the cost, outperforming most larger frontier models. It scores 54.9% on HLE-Verifi. In specialized domains requiring advanced analysis and reporting, 3.8 Flash surpasses 3.7 Flash and other frontier models on Vals Finance Agent V2 and Harvey's Legal Agent Benchmark. Google attributes these gains to one core design choice: 3.8 Flash 'works harder.' On complex tasks it executes extra reasoning steps and calls tools iteratively, and at higher effort levels it will even use more tokens to maximize performance. The blog also showcases compelling demos: building a fully playable DOS version of Google Maps from a single prompt in Google Antigravity, and an interactive 3D topographic visualizer of famous sites built from real U.S. Geological Survey datasets.

The other headline is Gemini 3.8 Flash Cyber — Google's most capable cybersecurity model to date. It ships through the new Fairwind Program to trusted government authorities, critical-infrastructure operators and software maintainers, delivering frontier-level vulnerability discovery and automated patching. On CyberGym, the standard industry benchmark for autonomous vulnerability discovery, 3.8 Flash Cyber surpasses both 3.5 Flash Cyber and significantly larger frontier models. On Google's comprehensive internal benchmark spanning complex codebases in 20 programming languages, it shows an impressive leap over previous models. Google stresses that it invested in vulnerability fixing from the start, prioritizing it over offensive capabilities like exploitation: on CWE-Bench, run by Collinear and widely recognized as a challenging external benchmark for patching, 3.8 Flash Cyber reaches the Pareto frontier with a pass@1 of 47.2%, just below a leading frontier model's 47.8% — at a significantly lower cost. Defenders get near-frontier patching at Flash-level pricing.

What is most telling is Google's own dogfooding results. The Chrome Security team found that 3.8 Flash Cyber produced 2.6 times more correct patches to Chrome vulnerabilities than the best commercial models that are much larger. Cloud security firm Wiz measured +7.5 to +9.7 percentage points higher recall on its internal penetration-testing benchmark versus other leading frontier models, at 2.3x to 5.2x lower cost. Google's Cloud Vulnerability Research team used 3.8 Flash Cyber to find a critical foundational vulnerability in under two hours — research and discovery that usually takes months. On safety design, 3.8 Flash ships with safeguards against misuse in CBRN (chemical, biological, radiological, nuclear) and cyber-offense domains per Google's Frontier Safety Framework; 3.8 Flash Cyber carries a more permissive set of mitigations for cybersecurity and is therefore limited to trusted defenders needing comprehensive cyber capabilities. According to Gray Swan, the 3.8 family also made a significant leap in prompt-injection robustness. On the consumer side, 3.8 Flash is available to Google AI Pro and Ultra subscribers across the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets.

For developers, enterprises and everyday users, this launch sends several clear signals. First, 'three Flashes in six weeks' shows Google is using the Flash line as its main weapon in an attrition war with rivals — every price-increase window is deliberately deferred, continuously raising the value-for-money bar with 'stronger at the same price.' Second, the Cyber variant follows a 'same model, different tiers' path similar to Anthropic's Mythos: wrapping the strongest capabilities into a dedicated program (Fairwind) for trusted defenders, addressing safety-regulator concerns while seeding Google's own security product ecosystem. Third, real-world numbers from Chrome, Wiz and the Cloud Vulnerability Research team show these models have evolved from 'good on lab benchmarks' to 'usable in real attack-defense work' — the human-efficiency ratio of the cybersecurity industry is being redefined. Worth watching next: which organizations gain access through Fairwind, the stability of 3.8 Flash in real long-running coding projects, and how OpenAI's GPT-5.x line responds to this dual-front competition on price and capability.

Placed in a longer timeline, Gemini 3.8 Flash means far more than 'another new model.' It proves that a low-price, high-volume product line can carry frontier capability: beating most larger frontier models on DeepSWE v1.1 and reaching a Pareto-frontier pass@1 of 47.2% on CWE-Bench, while pricing anchors below one dollar per million tokens — 'capability for everyone' and 'cost control' are no longer an either-or choice. For developers, this means attempting long-horizon coding and agentic tasks at Flash prices that previously required flagship models. For the security industry, 3.8 Flash Cyber's real-world results in the hands of Chrome, Wiz and Google Cloud's vulnerability research team herald an AI-driven defense-efficiency revolution. To be sure, reliability behind high-frequency iteration, Fairwind's access bar, and the chain reaction on rivals like OpenAI still need time to play out. But one thing is already clear: Google is sharpening the Flash line into one of the most cutting weapons in AI competition, on a cadence of one release every two weeks.