Gemini 3.8 Flash review: pricing, Cyber, and who should switch

Key takeaways

• Gemini 3.8 Flash launched September 2, 2026, three weeks after 3.7 Flash. Google’s third Flash drop in about six weeks.
• Positioned as Google’s best reasoning and coding Flash yet, at the same intro price as 3.7: $0.75 / 1M input and $3.75 / 1M output through December 31, 2026, then $1.50 / $7.50.
• Twin: Gemini 3.8 Flash Cyber for vulnerability discovery and patching, gated via the new Fairwind Program (trusted defenders / gov / critical infra).
• Available in Gemini API, AI Studio, Antigravity, Gemini Enterprise, and for Google AI Pro / Ultra in the Gemini app, AI Mode, and Sheets.
• Vendor benches favor long-horizon SWE and specialized agents. Treat Google’s charts as vendor harnesses, not independent audits.
• Pick 3.8 when agent diligence matters. Keep 3.7 (still supported) when you want fewer tokens and lower effort by default.

Table of contents

1. What shipped
2. Pricing and access
3. What Google claims improved
4. Flash vs Flash Cyber
5. Who should switch
6. FAQ

---

What shipped

Google did not ship a new Gemini Pro on September 2. It shipped another Flash iteration, and it is leaning hard into that cadence.

For practitioners:

Gemini 3.8 Flash is Google’s workhorse multimodal model for coding, agentic loops, and multi-step reasoning, priced like 3.7 Flash during the intro window. Gemini 3.8 Flash Cyber is the same foundation with cyber-focused capabilities, limited to Fairwind partners.

That separates:

• Gemini product surface: app, AI Mode, Sheets, Enterprise.
• 3.8 Flash (GA workhorse): API + Studio + Antigravity + consumer Pro/Ultra.
• 3.8 Flash Cyber (gated): Fairwind only. Not a public toggle for “better coding.”

Knowledge cutoff framing (model card / coverage): March 2026 for some domains; others may still feel closer to January 2025. Confirm against the live model card before you rely on recent facts.

Pricing and access

Verified September 2026 against Google’s announcement.

	Gemini 3.8 Flash	Gemini 3.8 Flash Cyber
Role	GA workhorse Flash	Cyberdefense specialist
API intro (through Dec 31, 2026)	$0.75 in / $3.75 out per 1M tokens	Fairwind program pricing / access
API after Jan 1, 2027	$1.50 in / $7.50 out	Program-dependent
Developer surfaces	Gemini API, AI Studio, Antigravity, Android Studio, Stitch	Fairwind partners
Consumer	Google AI Pro / Ultra (Gemini app, AI Mode, Sheets)	Not a consumer SKU
Enterprise	Gemini Enterprise	Fairwind for trusted defenders

Introductory pricing is the story. If your agents burn a lot of tokens, the Dec 31 cliff matters more than the marketing charts.

What Google claims improved

From Google’s post (vendor numbers):

• Strong showing on DeepSWE v1.1 long-horizon software engineering versus many larger frontier models at Flash cost.
• Gains on specialized agent benches such as Vals Finance Agent V2 and Harvey Legal Agent Benchmark versus 3.7 Flash.
• HLE-Verified 54.9% cited for multi-step reasoning across STEM and professional fields.
• Design note: 3.8 “works harder” (more reasoning steps and tool calls on complex tasks). Lower effort settings or staying on 3.7 remain options when token burn is the constraint.
• Safety: CBRN and cyber-offense safeguards on standard Flash; Cyber is more permissive for defensive cyber and therefore gated. Prompt-injection robustness gains claimed via Gray Swan measurements.

Demo narratives lean Antigravity builds (games, DOS Maps, USGS topo, Three.js teardowns). Fun for marketing; not a substitute for your repo evals.

Flash vs Flash Cyber

	3.8 Flash	3.8 Flash Cyber
Audience	Builders and Pro/Ultra users	Trusted defenders via Fairwind
Focus	Coding, agents, general multi-step work	Vulnerability discovery + automated patching
Safeguards	Standard Flash / Frontier Safety Framework	More permissive cyber mitigations
Public access	Yes	Application / vetting

Google cites CyberGym frontier-level discovery, >70% on an internal multi-language vuln bench, CWE-Bench pass@1 about 47.2% near a larger frontier model at higher cost, Chrome Security 2.6× more correct patches than best commercial models tested, and partner notes from Wiz and Cloud Vuln Research. Again: vendor and partner claims.

Who should switch

Move to 3.8 Flash when:

• You already like Gemini Flash economics and want better long-horizon coding agents.
• You live in Antigravity / AI Studio and want the newest workhorse.
• You are comparing Flash-class cost to Claude Opus/Fable or ChatGPT frontier SKUs and need a cheaper agent loop.

Stay on 3.7 (or lower effort) when:

• Token burn and latency dominate, and 3.8’s “works harder” behavior hurts unit economics.
• Your evals do not move on your real tasks.

Apply to Fairwind when: you are a defender org that needs Cyber’s vuln/patch loop. Everyone else should ignore Cyber marketing as a public upgrade path.

Also compare the same-day Meta drop: Muse Spark 1.3 review and Gemini 3.8 Flash vs Muse Spark 1.3.