Google ships Gemini 3.8 Flash and Flash Cyber at 3.7 Flash prices

Google introduced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber on 2 September 2026. Raluca Ada Popa, Gemini security lead at Google DeepMind, wrote the announcement, which calls 3.8 Google's best reasoning and coding model so far at the speed and price of 3.7 Flash, and its third Flash release in six weeks.

Gemini 3.8 Flash is Google's workhorse model for agentic coding and multi-step reasoning, priced at $0.75 per million input tokens and $3.75 per million output tokens until 31 December 2026. Gemini 3.8 Flash Cyber is the same core trained for vulnerability discovery and patching, available only to trusted defenders through Google's new Fairwind Program.

What changed

• Same price as 3.7 Flash, for now. The introductory rate is $0.75 per million input tokens and $3.75 per million output tokens. From 1 January 2027 the price rises to $1.50 and $7.50.
• Long-horizon coding. On DeepSWE v1.1, Google says 3.8 Flash outperforms most larger frontier models at solving engineering problems end to end, at a fraction of their cost. It scores 54.9% on HLE-Verified and beats 3.7 Flash and other frontier models on Vals Finance Agent V2 and on Harvey's Legal Agent Benchmark.
• It spends more tokens to get there. Google is explicit that 3.8 Flash "works harder": more reasoning steps and more iterative tool calls, especially at higher effort levels. Teams whose main constraint is compute are told to use lower effort levels or stay on 3.7 Flash, which remains supported.
• Flash Cyber's numbers. Frontier-level pass@1 on CyberGym, above 3.5 Flash Cyber and above larger models. On an internal benchmark spanning 20 programming languages it finds vulnerabilities with a success rate above 70%. On Collinear's CWE-Bench for patching, it scores 47.2% pass@1 against 47.8% for a leading frontier model, at much lower cost.
• Inside Google. The Chrome Security team found 3.8 Flash Cyber produced 2.6 times more correct patches to Chrome vulnerabilities than the best commercial models. Wiz measured 7.5 to 9.7 points higher recall on its internal penetration-testing benchmark at 2.3 to 5.2 times lower cost. Google's Cloud Vulnerability Research team used it to find a critical vulnerability in under two hours.
• Safeguards. 3.8 Flash ships with CBRN and cyber-offence mitigations under Google's Frontier Safety Framework. Flash Cyber carries a more permissive cyber policy, which is why access is gated. Google says both 3.8 models made a significant jump in prompt injection resistance as measured by Gray Swan.
• Where to get it. Google Antigravity, the Gemini API through Google AI Studio, Android Studio, Stitch, Gemini Enterprise, and the Gemini app for Google AI Pro and Ultra subscribers. Flash Cyber requires a Fairwind Program application.

Why it matters for agent builders

Flash is the tier most production agents actually run on, because a loop that calls a model dozens of times per task cannot afford frontier prices on every call. Google's pitch is that 3.8 Flash closes much of the gap to those frontier models on agentic benchmarks without moving off the Flash price, at least until the introductory rate expires.

The token caveat matters as much as the price. A model that reasons longer and calls tools more often can cost more per task than its predecessor even at the same per-token rate. Google's guidance to drop the effort level for cost-sensitive workloads is the knob to test before switching a fleet of agents over.

The prompt injection result is the other line worth reading twice. Every agent that reads web pages or email is exposed to injected instructions, and Google reports a large improvement for the Gemini 3.8 models on Gray Swan's indirect prompt injection benchmark. The coding agents on AI Agents Listing that expose a model picker will be the first place most developers try it.

What to watch

The price doubles on 1 January 2027, so any cost model built on $0.75 per million has four months of shelf life. The other open question is how much of the agentic gain survives at the lower effort levels Google recommends for efficiency: the announcement reports headline scores without saying which effort setting produced them.

Key takeaways

• Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens until 31 December 2026, then $1.50 and $7.50.
• Google says 3.8 Flash beats most larger frontier models on DeepSWE v1.1 and scores 54.9% on HLE-Verified.
• 3.8 Flash uses more reasoning steps and tool calls than 3.7 Flash, so per-task cost can rise at the same token price.
• Gemini 3.8 Flash Cyber scores 47.2% pass@1 on CWE-Bench and is available only through the Fairwind Program.
• The Chrome Security team measured 2.6 times more correct vulnerability patches from Flash Cyber than from larger commercial models.

FAQ

How much does Gemini 3.8 Flash cost?

$0.75 per million input tokens and $3.75 per million output tokens, the same introductory price as Gemini 3.7 Flash. The introductory price expires on 31 December 2026, after which it is $1.50 and $7.50.

What is Gemini 3.8 Flash Cyber?

A variant of 3.8 Flash trained for vulnerability discovery and automated patching, with a more permissive set of cybersecurity mitigations. Google offers it only to trusted defenders, including government authorities, critical infrastructure operators and software maintainers, through the Fairwind Program.

Should I move my agents from 3.7 Flash to 3.8 Flash?

Google says 3.8 Flash is more capable on agentic and coding tasks at the same price, and also that it uses more tokens on complex tasks. Measure cost per completed task at the effort level you plan to run, and keep 3.7 Flash in mind for efficiency-first workloads; it remains fully supported.

Where can I use Gemini 3.8 Flash today?

Developers can use it in Google Antigravity, through the Gemini API in Google AI Studio, in Android Studio and in Stitch. Enterprises get it in Gemini Enterprise, and consumers on Google AI Pro and Ultra plans get it in the Gemini app, AI Mode in Search and Google Sheets.

Source: Google Discussion: Hacker News

The Coding category is where to compare the agents that let you pick a model per task.