Summary: Google released Gemini 3.8 Flash and Gemini 3.8 Flash Cyber on September 2, 2026, three weeks after 3.7 Flash. 3.8 Flash reports 54.9% on HLE-Verified, with introductory prices doubling from $0.75/$3.75 to $1.50/$7.50 per million input/output tokens on January 1, 2027. Flash Cyber is restricted through the Fairwind Program and reports 47.2% pass@1 on CWE-Bench.
On September 2, 2026, Google shipped two models at once: Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. Only three weeks separate them from 3.7 Flash, which landed on August 13, and Google’s own words call this its “third Flash model release in six weeks.” The direction is unmistakable — Pro-tier updates have decelerated while Flash keeps shipping, and the company is putting its weight behind the fast, cheap, smart workhorse tier. The model card DeepMind published alongside the release sets the knowledge cutoff at March 2026.
One core intelligence sits underneath both variants. The official post carries the names of Tulsee Doshi and Raluca Ada Popa, and it credits this generation’s improvement to “long-running agentic loops designed to recursively evaluate and refine the underlying models,” with a further share of the gains traced to “rigorous training in the highly demanding domain of cybersecurity.”
3.8 Flash Benchmark Results: A Workhorse That Trades Punches With Frontier Models
Google bills 3.8 Flash as its “most intelligent workhorse model,” beating 3.7 Flash across software engineering, agentic tasks, and multi-step specialized reasoning — without giving up any speed or cost. The figures:
- On DeepSWE v1.1, a long-horizon software engineering benchmark, it comes out ahead of most larger frontier models
- It wins against 3.7 Flash and other frontier models on Vals Finance Agent V2 and Harvey’s Legal Agent Benchmark
- HLE-Verified comes in at 54.9%
The real weight of those scores is in the cost structure: frontier-adjacent results at Flash-tier prices. For a team pushing agentic workloads at scale, that counts for more than any single peak benchmark.
Pricing and the “Works Harder” Tradeoff
3.8 Flash launches on introductory rates — $0.75 per million input tokens and $3.75 per million output tokens — good through December 31, 2026. From January 1, 2027, those figures become $1.50 and $7.50. A team that wants to lock pricing in for the cheap window should work the numbers now.
Google is also unusually direct about an engineering tradeoff: 3.8 Flash “works harder,” spending more reasoning steps and tool calls when effort is turned up, which pushes token consumption higher. Where the priority is efficiency rather than peak intelligence, Google points users to lower effort settings or to staying on 3.7 Flash. A lab stating plainly that its new model is not the cheaper option everywhere is a rare thing.
Flash Cyber: Defense as Its Own Model
Flash Cyber ranks as Google’s strongest cybersecurity model, marketed on “frontier-level performance in vulnerability detection and automated patching” and built for defense instead of offense. Its headline results:
- It tops 3.5 Flash Cyber and larger frontier models on CyberGym
- An internal benchmark spanning 20 programming languages puts its success rate above 70%
- CWE-Bench (Collinear) comes in at 47.2% pass@1, essentially level with a leading frontier model’s 47.8% while costing substantially less
- Chrome Security recorded 2.6x as many correct patches as larger commercial models
- Wiz measured 7.5–9.7 percentage points more recall at 2.3–5.2x lower cost
- Google’s own Cloud Vulnerability Research team located a critical foundational vulnerability in under 2 hours — finds of that kind normally take months
Since Flash Cyber carries more permissive cyber-safety mitigations, general developers cannot get it. Access runs through the Fairwind Program, prioritized for trusted government authorities, critical infrastructure operators, and software maintainers.
Availability
3.8 Flash covers broad ground: developers find it in Google Antigravity, the Gemini API (AI Studio and Android Studio), and Stitch; enterprises reach it through Gemini Enterprise; consumers see it roll out via AI Pro/Ultra subscriptions in the Gemini app, AI Mode in Search, and Gemini in Sheets. Safeguards continue under Google’s Frontier Safety Framework, with prompt-injection robustness strengthened based on Gray Swan testing.
For product teams, the message lands clearly: Flash has become Google’s primary vehicle for iteration, and cyber defense now stands as a product line of its own rather than a feature attached to the general-purpose model.