Gemini 3.8 Flash vs. Flash Cyber: What’s Different?

Google introduced Gemini 3.8 Flash on September 2, 2026, in two versions: a general-purpose model for coding, agent workflows and reasoning, and Flash Cyber, a cybersecurity-focused model for finding and patching vulnerabilities. The distinction is practical as well as technical: standard Flash has published API pricing, while Cyber is available through Google’s restricted Fairwind Program for trusted defenders. 824

Why 3.8 Flash followed so quickly

Google described the release as its third Flash model in six weeks. Gemini 3.7 Flash had arrived about three weeks earlier, according to launch coverage, making 3.8 a rapid follow-up rather than a model announced after a long gap. 46

Google positioned 3.8 Flash as its general-purpose “workhorse” model, with improvements over 3.7 in software engineering, agentic tasks and multi-step reasoning. 8 That makes it the broadly relevant option for developers and teams evaluating coding or agent-based workflows; the benchmark results below offer a snapshot, not a guarantee for every workload.

What the reported benchmarks show

For the standard model, Google’s reported results include 90.8% on Terminal-Bench 2.1, compared with 81.6% for Gemini 3.7 Flash, and 54.9% on HLE-Verified, a test of multi-step reasoning. 3 Those numbers indicate gains on the cited evaluations, but they should not be read as a single overall measure of model quality.

Flash Cyber targets a different set of tasks. Reported results include 86.2% on CyberGym for vulnerability discovery and 47.2% on CWE-Bench for patching. Google also reported a success rate above 70% on an internal vulnerability-discovery evaluation spanning 20 programming languages. 715

These results are not directly comparable: CyberGym, CWE-Bench and the internal evaluation test different tasks and use different evaluation setups. They are evidence about specific tests, not proof that Cyber will find or fix vulnerabilities reliably in any organization’s codebase.

Pricing: standard Flash starts at the 3.7 rate

Gemini 3.8 Flash launched at the same introductory API rates as 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens. 8 Google DeepMind’s model card also lists regular rates of $1.50 per million input tokens and $7.50 per million output tokens; launch coverage reports that those higher rates take effect on January 1, 2027. 110

The published figures here concern standard Flash. The available sources do not establish a public API price for Flash Cyber.

Flash Cyber is gated through Fairwind

Flash Cyber is designed for cybersecurity work, including vulnerability discovery and automated patching. Google’s Fairwind Program is a limited-access program for governments and trusted partners to use its advanced cyber-defense capabilities. Google says Fairwind combines Flash Cyber with its CodeMender harness to help defenders find, verify and fix vulnerabilities. 24

Unlike a generally available model, Flash Cyber is limited to approved defenders. Reporting describes its cyber-specific safeguards as more permissive than those on the standard model, while the Fairwind restriction limits who can use those capabilities. 2128 The access model reflects the sensitivity of tools that can discover and help remediate software vulnerabilities; it also means most developers cannot evaluate Flash Cyber simply by signing up for the standard API.

Which version should teams consider?

For general coding, agent workflows and reasoning, Gemini 3.8 Flash is the broadly accessible option, with published API rates and reported improvements on selected benchmarks. 38 Flash Cyber is aimed at organizations conducting defensive security work, but access depends on Fairwind eligibility; its benchmark claims should be assessed against the specific vulnerability-discovery or patching task at hand. 724

The clearest takeaway is that these are related models with different intended uses and access boundaries—not interchangeable products. Standard Flash offers a general-purpose upgrade path; Cyber packages security-focused capabilities behind a restricted program.