Gemini 3.8 Flash arrives at 3.7's introductory price, with Cyber access restricted (2026)

Google DeepMind announced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber on September 2, 2026. The company calls Gemini 3.8 Flash its strongest reasoning and coding model so far, built on the Gemini 3.7 Flash core at the same introductory price. Gemini 3.8 Flash is generally available; Gemini 3.8 Flash Cyber is reserved for trusted defenders through the Fairwind Program. 1

What launched

Signal	Confirmed detail	Why it matters	Action window
Shared foundation	Gemini 3.8 Flash is based on Gemini 3.7 Flash. It accepts text, images, audio, and video in a context window of up to 1 million tokens, with up to 64K output tokens. Effort levels trade quality, cost, and latency. 2	The upgrade targets long tasks and agents.	Replay long-context and tool-use tasks at the same effort level.
Coding and agents	Google reports 73.7% on DeepSWE v1.1, 61.4% on Vals Finance Agent v2, 10.0% on Harvey's Legal Agent Benchmark, and 54.9% on HLE-Verified. 1	The target is work requiring repeated tool calls, verification, and domain context.	Compare task completion, wall-clock time, and token use.
Price and distribution	The introductory price is $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. Google lists the Gemini API, AI Studio, Gemini Enterprise, Google Antigravity, the Gemini app, and AI Mode as access points. 1 2	The price supports high-volume experiments, while the deadline changes longer-term estimates.	Budget for $1.50/$7.50 per million tokens from January 1, 2027.
Cyber variant	Fairwind provides Gemini 3.8 Flash Cyber to trusted government authorities, critical-infrastructure operators, and software maintainers. Google reports 47.2% pass@1 on CWE-Bench and more than 70% on an internal vulnerability-discovery test across 20 programming languages. 1	The defensive-security capability has controlled access.	Eligible defenders can apply; other teams can track the results without assuming access.

What the numbers say

Google's comparison puts Gemini 3.8 Flash ahead of Gemini 3.7 Flash on the listed coding, knowledge-work, and reasoning tests. Claude Opus 5 leads several entries, including DeepSWE v1.1 and Terminal-bench 4.0. The release therefore combines a broad company-reported evaluation lead with performance that still varies by workload. 2

Google DeepMind's benchmark comparison for Gemini 3.8 Flash — *Google DeepMind's September 2026 model-card comparison, including prices and selected benchmark results reported by Google. 2*

Limits that affect deployment

The model card lists hallucinations, occasional slowness or timeouts, and higher token use at higher effort levels. The knowledge cutoff is March 2026, with some domains limited to January 2025. Google's safety table reports broadly similar performance to Gemini 3.7 Flash alongside a slight multilingual-safety regression. Revised automated evaluations are not directly comparable with earlier model-card results. 2

Why this matters

Gemini 3.8 Flash is a low-cost candidate for long-horizon coding and agent workflows, with enough context for large repositories and documents. Gemini 3.8 Flash Cyber is a separate, restricted defensive track rather than a generally available API model.

Replay production-like tasks against Gemini 3.7 Flash and 3.8 Flash. Compare successful completion, latency, retries, and total tokens before changing model routing. Eligible security teams can pursue Fairwind access; everyone else can treat the Cyber results as a signal for future evaluation.