Gemini 3.8 Flash: Unchanged Prices, Filtered Cyber Access

Key Takeaways

1Google launches Gemini 3.8 Flash for coding and agents, and Flash Cyber for cybersecurity

2Flash Cyber is only accessible to trusted defenders via Fairwind

3Internal teams and Wiz report measurable gains in vulnerability detection and remediation

4Generation 3.8 shows improved robustness against prompt injection

Why it matters : Gemini 3.8 aims to enhance software autonomy and cyber defense while maintaining competitive pricing and security safeguards.

Google launches Gemini 3.8 in two versions: Flash, focused on coding and agents, and Flash Cyber, dedicated to cybersecurity. Access to Flash Cyber is restricted to trusted defenders through the Fairwind program. Internal teams and partners report quantified gains, while Flash maintains the introductory price of 3.7.

Chrome, Wiz, and Cloud Teams Report Quantified Gains

The Chrome security team has observed that Gemini 3.8 Flash Cyber produces 2.6 times more valid patches for Chrome vulnerabilities than some significantly larger commercial models. Wiz reports achieving a recall improvement of +7.5 to +9.7% on its own penetration testing benchmark, while displaying a cost 2.3 to 5.2 times lower than other benchmark models. The cloud vulnerability research team at Google used 3.8 Flash Cyber to identify a fundamental critical vulnerability in less than 2 hours, whereas such discoveries typically take several months.

Priority Defense and Restricted Access via Fairwind

Gemini 3.8 Flash Cyber is reserved for trusted defenders through the Fairwind program, with a focus on defensive advantage and rapid iteration. The development of the model emphasized automated vulnerability remediation rather than offensive capabilities like exploitation. Flash Cyber incorporates more permissive security measures for cybersecurity and is only accessible to entities requiring advanced capabilities, such as government authorities, critical infrastructure operators, and software maintainers, upon application.

Cyber Performance on Public and Internal Benchmarks

On the CyberGym benchmark, a reference for vulnerability detection, Gemini 3.8 Flash Cyber demonstrates advanced-level performance and surpasses 3.5 Flash Cyber as well as much larger models. An internal assessment covering 20 programming languages reveals a success rate exceeding 70%. On CWE-Bench, administered by Collinear, the model achieves a 47.2% success rate, a score close to that of a benchmark model at 47.8%, while being offered at a significantly lower cost.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Flash 3.8: Agents, Long-Term Coding, and Multi-Step Reasoning

Gemini 3.8 Flash is designed for long-term coding and autonomous agents. On DeepSWE v1.1, it outperforms the majority of larger models in autonomously solving complex end-to-end problems, while displaying a much lower cost. It is considered reliable enough for critical enterprise autonomy in specialized sectors and outperforms 3.7 Flash as well as other models on Vals Finance Agent V2 and Harvey’s Legal Agent Benchmark. It achieves 54.9% on HLE-Verified, highlighting its multi-step reasoning capabilities in STEM, humanities, and professional domains. The model can perform more reasoning steps and tool calls, consuming more tokens at higher effort levels; lower effort levels are offered when efficiency is prioritized, and 3.7 Flash remains supported for these uses.

Same Pricing as 3.7 and Multi-Channel Availability

Gemini 3.8 Flash retains the introductory price of 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens. Developers can create agent flows in Google Antigravity, use the Gemini API via Google AI Studio and Android Studio, and generate user interfaces in Stitch. Businesses access it through Gemini Enterprise. For consumers subscribed to Google AI Pro and Ultra, access is through the Gemini app, Google Search's AI mode, and Gemini in Google Sheets.

Safety Framework and Increased Robustness Against Prompt Injection

Gemini 3.8 Flash includes safety measures aimed at preventing misuse in chemical, biological, radiological, and nuclear domains as well as for offensive applications, in accordance with the Frontier security framework. According to Gray Swan assessments, generation 3.8 shows significant progress in robustness against prompt injection attacks to protect users. Both variants rely on the same intelligence base, enhanced by long-duration agent loops and rigorous training including cybersecurity. Gemini 3.8 is presented as the most advanced reasoning and coding model from the publisher, while maintaining the speed and reduced cost of 3.7, and is available in Flash and Flash Cyber; the latter positioned as the most effective in cybersecurity.