Gemini 3.8 Flash: What's New for Coding and AI Agents

Google just shipped Gemini 3.8 Flash, its third Flash release in roughly six weeks and its "most intelligent workhorse model" yet. The pitch is simple: meaningful gains in software engineering, agentic tasks, and multi-step reasoning — at the same speed and low price as 3.7 Flash. A separate Gemini 3.8 Flash Cyber variant targets vulnerability discovery and patching. This guide covers what changed, the benchmarks Google published, what it costs, and where it fits if you build AI agents.

What is Gemini 3.8 Flash?

Gemini 3.8 Flash is a lightweight, multimodal model in Google's Flash line, tuned for long-horizon coding and autonomous agents. Google frames it as "our most intelligent workhorse model," delivering improvements over 3.7 Flash across engineering, agentic work, and reasoning in specialized domains (Google).

It launched on September 2, 2026, three weeks after 3.7 Flash — part of a cadence where Google has been shipping a new Flash model roughly every three weeks since Gemini 3.6 Flash in late July (Thurrott).

The release includes two variants built on the same foundational intelligence:

• Gemini 3.8 Flash — the general workhorse model for coding, agents, and enterprise workflows.
• Gemini 3.8 Flash Cyber — a cybersecurity-focused model for vulnerability detection and automated patching, available only to trusted defenders through a new Fairwind Program.

What's new versus Gemini 3.7 Flash

The headline is that 3.8 Flash "delivers substantial gains" over 3.7 Flash while "often approaching the performance of higher-cost frontier models" — without raising the price (9to5Google).

Google attributes the jump to one design choice: the model works harder. On complex tasks it does more reasoning steps and calls tools iteratively, which can mean higher token usage at higher effort levels. If compute cost is your main constraint, Google says you can dial down the effort level or stay on 3.7 Flash, which remains supported for efficiency-first workloads.

That trade-off matters for agent builders. More diligence and iterative tool-calling is exactly what long-running agent loops need — but it also means you should watch token spend on high-volume, low-complexity tasks.

Gemini 3.8 Flash benchmarks

Google published results across coding, finance, legal, and general reasoning. All figures below come from Google's launch materials, so read them as vendor benchmarks rather than independent tests.

• DeepSWE v1.1 (Long-Horizon Software Engineering): 3.8 Flash outperforms most larger frontier models at autonomously solving complex engineering problems end to end, at a fraction of the cost (Google).
• Vals Finance Agent V2 and Harvey's Legal Agent Benchmark: 3.8 Flash beats 3.7 Flash and other frontier models in these specialized professional domains.
• HLE-Verified (Humanity's Last Exam): 54.9%, which Google cites as evidence of multi-step reasoning across STEM, humanities, and professional fields (9to5Google).

The through-line is agentic capability in domains where a model has to plan, use tools, and stay on task across many steps — not just answer a single prompt.

One caveat worth noting: like the previous model, Gemini 3.8 Flash's knowledge cutoff is March 2026 for some domains, while in others its knowledge may be limited to January 2025 (9to5Google). Plan to give it fresh context or tools for anything recent.

Gemini 3.8 Flash pricing

Google kept the introductory price flat with 3.7 Flash:

• $0.75 per million input tokens
• $3.75 per million output tokens

That introductory rate runs until December 31, 2026; after that the standard price rises to $1.50 per million input and $7.50 per million output tokens (Google). If you're building on it, factor the January price change into any long-term cost model — and remember that the "works harder" behavior can push output token counts up on complex tasks.

Where to use Gemini 3.8 Flash

Google made it available across consumer, developer, and enterprise surfaces on day one:

• Consumers: the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets, for Google AI Pro and Ultra subscribers (Search Engine Journal).
• Developers: Google Antigravity, AI Studio, Android Studio, and the Gemini API.
• Enterprises: Gemini Enterprise.

For agent developers, API access is the important one — it lets you slot 3.8 Flash into your own orchestration rather than Google's apps.

Gemini 3.8 Flash Cyber: the security variant

The Cyber variant is the more novel part of this launch. It's aimed at defenders and prioritizes vulnerability fixing over offensive capabilities. Google reports frontier-level performance on CyberGym, the standard benchmark for autonomous vulnerability discovery, and a success rate above 70% on an internal benchmark spanning 20 programming languages (Google).

Real-world results Google cites from its own security teams:

• Chrome's Security team found 3.8 Flash Cyber produced 2.6x more correct patches to Chrome vulnerabilities than much larger commercial models (9to5Google).
• Security vendor Wiz measured 7.5–9.7% higher recall on an internal penetration-testing benchmark at 2.3–5.2x lower cost than other leading frontier models.
• Google's Cloud Vulnerability Research team used it to find a critical foundational vulnerability in under two hours — work that usually takes months.

The catch: 3.8 Flash Cyber is gated behind the Fairwind Program for trusted governments, critical-infrastructure operators, and software maintainers. It isn't a general API release.

Should you switch from 3.7 Flash?

A quick decision guide based on what Google published:

• Building coding or multi-step agents? 3.8 Flash is the upgrade — better long-horizon engineering and tool use are the whole point.
• Running high-volume, simple tasks where cost dominates? Stay on 3.7 Flash or use a lower effort level; the "works harder" behavior can raise token bills.
• Defensive security work and eligible for Fairwind? Apply for 3.8 Flash Cyber.
• Need current knowledge? Wire in retrieval or live tools — the cutoff means the model won't know recent events on its own.

Do this with your own AI workforce

Gemini 3.8 Flash is built for exactly the pattern Eigent runs on: long-horizon, tool-calling agents that plan and execute real work instead of answering one prompt. Eigent is an open-source "Cowork" desktop app — a multi-agent workforce that automates workflows locally, so you can point capable models at coding, research, and document tasks and keep the data on your machine. If you're eyeing the agentic gains in 3.8 Flash, see how reviewing GitHub PRs works with an agent workforce, or compare notes with our breakdown of the best open-source AI coding agents. Download Eigent to build your own.