Google Gemini 3.8 Flash and Flash Cyber Announced: Everything New
Originally published in Turkish. Read the original →
Google has released the Gemini 3.8 Flash and Flash Cyber models. All the details on multi-step reasoning, software engineering, the unchanged launch pricing and the cybersecurity-focused Flash Cyber version.
In the race between AI models, launch schedules are now measured in weeks rather than months. Only three weeks after releasing Gemini 3.7 Flash, which made a big splash in the developer world, Google has officially rolled out the next-generation Gemini 3.8 Flash and the cybersecurity-focused Gemini 3.8 Flash Cyber.
This release stands out as the third major Flash model update Google has shipped in the last three months. Taking the "fast and economical" label much further, the 3.8 series offers a capability set that competes directly with the industry's most expensive flagship models in software engineering, multi-step reasoning and agentic AI.
Multi-Step Reasoning and a Model Architecture That "Works Harder"
While traditional fast models focus on processing input tokens with minimal latency and producing an answer directly, Gemini 3.8 Flash takes a different approach. While positioning the model as one of its most advanced general-purpose models, Google enabled it to run additional reasoning steps on difficult tasks in particular and to use external tools (tool calling) iteratively, one after another.
- The main consequences of this architectural choice are:
- In-Depth Problem Solving: Instead of answering in a single pass, the model forms intermediate hypotheses, runs code, checks the output and, if necessary, corrects its course before reaching the final answer.
- Dynamic Effort and Token Consumption: At high effort levels, the model can spend more thinking tokens; in return, it achieves accuracy at the level of frontier models.
- Agent Capability: Its ability to chain together API calls, file system searches and terminal commands makes 3.8 Flash one of the strongest candidates in its class for autonomous AI agents.
A Leap in Software Engineering: DeepSWE v1.1 and Autonomous Coding
The area developers are most curious about is undoubtedly coding performance. According to data shared by Google, Gemini 3.8 Flash showed a dramatic performance increase over Gemini 3.7 Flash on DeepSWE v1.1, a benchmark that tests long-running software engineering workloads.
The model managed to outperform many huge and costly models in the industry on end-to-end autonomous engineering processes such as finding bugs in complex repos, architectural refactoring and building features from scratch. The context mastery the model offers in developer environments (IDEs, terminal agents and CI/CD automation) significantly reduces software teams' routine maintenance load.
Top-Level Results in Finance, Law and HLE-Verified
Gemini 3.8 Flash's capabilities are not limited to software development. The model also posted notable scores on enterprise tests that require deep domain expertise and rigorous analysis:
- Finance and Data Analysis (Vals Finance Agent V2): It outperformed Gemini 3.7 Flash and other top-tier models in its class on financial reporting, balance sheet analysis and modeling work.
- Legal Reasoning (Harvey's Legal Agent Benchmark): It showed high consistency in reviewing legislation, auditing contracts and extracting precedents.
- Humanity's Last Exam (HLE-Verified): It scored 54.9% on the HLE-Verified benchmark, which measures the hardest multi-step reasoning questions across STEM, the humanities and professional disciplines.
These figures show that the Flash family is turning from a mere "helper assistant" into an "expert agent" layer that directly runs complex business processes.
Pricing Unchanged: The Attractive Developer Rate Continues
Despite the performance leap the new model brings, Google kept its developer-friendly pricing policy. Under the launch pricing valid until the end of the year, Gemini 3.8 Flash's token costs have not changed:
| Pricing Item | Price (Per 1 Million Tokens) | Valid Until |
|---|---|---|
| Input Tokens | $0.75 | December 31 |
| Output Tokens | $3.75 | December 31 |
Gemini 3.8 Flash Cyber: A New Standard in Autonomous Cybersecurity
One of the most strategic items in Google's announcement was Gemini 3.8 Flash Cyber. Designed as the successor to the previously released 3.5 Cyber model, this version is specialized for vulnerability research, attack surface analysis and autonomous patch development.
A Trusted Testing Process with the Fairwind Program
To minimize the risk of misuse, Flash Cyber will not be opened to the general public at first. The model is offered only to security researchers and approved enterprise testers under Google's new Fairwind Program.Critical Security Tests and Wiz Results
The first security metrics Google announced make the model's defense-oriented capabilities concrete: 1. Chrome Security Success: According to data shared by the Google Chrome Security team, 3.8 Flash Cyber produced 2.6 times more correct patches for complex browser vulnerabilities than the industry's major commercial models. 2. Wiz Penetration Test Data: In internal penetration tests run by leading cloud security company Wiz, the model showed 7.5% to 9.7% higher vulnerability detection success than other leading AI models. 3. Cost Advantage: According to Wiz's report, this high security performance was achieved at 2.3 to 5.2 times lower operational cost than rival frontier systems.Frequently Asked Questions (FAQ)
When was Gemini 3.8 Flash released?
Google opened the Gemini 3.8 Flash and Gemini 3.8 Flash Cyber models to developers and partners as of September 3, 2026.How is Gemini 3.8 Flash different from previous versions?
The model stands out for its depth of multi-step reasoning and its ability to use external tools in a loop. On complex benchmarks such as DeepSWE v1.1, Vals Finance Agent and Harvey Legal, it delivers a clear performance increase over previous Flash models and large commercial rivals.Is Gemini 3.8 Flash Cyber generally available?
No. Because the Flash Cyber version is designed specifically for vulnerability research and patch generation, it is offered to approved security researchers and organizations through Google's controlled Fairwind Program.What is the model's API pricing?
The launch price for developers is $0.75 per 1 million input tokens and $3.75 per 1 million output tokens. This rate is valid until December 31.Overall Assessment
Google's Gemini 3.8 Flash move proves once again that in the AI industry, increasing reasoning efficiency is just as critical as scaling up models. The model's ability to take extra thinking rounds at high effort levels takes autonomy in software engineering tasks to the next level. Flash Cyber, positioned on the cyber defense side, sends very strong signals about the future of security automation with its Wiz and Chrome Security results.
