The news
On September 2, 2026, Google released Gemini 3.8 Flash, a low-cost AI model it called its strongest for reasoning and coding so far, alongside Gemini 3.8 Flash Cyber, a security-tuned version available only to vetted defenders through a new Fairwind Program.
Google said the launch came three weeks after Gemini 3.7 Flash and was its third Flash release in six weeks. Tech site Thurrott reported the three-week cadence dated to July's 3.6 Flash. Gemini 3.8 Flash carries the same introductory price as 3.7 Flash: $0.75 per million input tokens and $3.75 per million output tokens, until December 31, 2026. From January 1, 2027, the price rises to $1.50 and $7.50. Tokens are the chunks of text a model reads and writes.
Google said 3.8 Flash outperforms most larger frontier models on DeepSWE v1.1, a long-horizon software engineering benchmark, and beats 3.7 Flash and other frontier models on Vals Finance Agent V2 and Harvey's Legal Agent Benchmark. It reported a 54.9% score on HLE-Verified. The company said the model takes extra reasoning steps and calls tools repeatedly, so it may use more tokens at higher effort settings; developers can lower the effort level or stay on 3.7 Flash, which remains fully supported.
Developers can use 3.8 Flash through the Gemini API in Google AI Studio and Android Studio, and in Google Antigravity and Stitch. Businesses get it in Gemini Enterprise. Google AI Pro and Ultra subscribers get it in the Gemini app, AI Mode in Google Search and Gemini in Google Sheets.
For the Cyber model, Google said it beat 3.5 Flash Cyber and much larger models on the CyberGym vulnerability-finding benchmark, topped 70% success on an internal test spanning 20 programming languages, and scored 47.2% on the CWE-Bench patching test against 47.8% for a leading frontier model. Google said its Chrome Security team found the model produced 2.6 times more correct patches for Chrome vulnerabilities than the best commercial models. Google said it prioritized fixing flaws over offensive skills such as exploitation.
In a separate post, Google vice president Four Flynn said Fairwind pairs 3.8 Flash Cyber with CodeMender, Google's tool for writing and validating code fixes. Access is staged for governments and national cyber authorities, critical infrastructure operators and core technology platforms, which must limit use to internal security, incident response or penetration testing staff and deploy protections such as multi-factor authentication. Google said the program has more than 650 participating partners.
The numbers
- Introductory price, input (through Dec. 31, 2026)
- $0.75 per million tokens
- Introductory price, output (through Dec. 31, 2026)
- $3.75 per million tokens
- Price from Jan. 1, 2027 (input / output)
- $1.50 / $7.50 per million tokens
- HLE-Verified score (Google)
- 54.9%
- CWE-Bench patching, 3.8 Flash Cyber (Google)
- 47.2% pass@1
- Fairwind participating partners (Google)
- More than 650
Why CEOs should care
For technology buyers, the headline price is temporary. Budgets built on $0.75 and $3.75 per million tokens need a line for the January 1, 2027 step to $1.50 and $7.50. Google also said 3.8 Flash may spend more tokens to do better on hard tasks, so compare cost per completed task, not cost per token, and ask whether the cheaper 3.7 Flash or a lower effort setting is good enough for each workload.
CISOs should read the Cyber results as vendor claims until proven on their own code. The Chrome result comes from Google's own security team, and the other figures were published by Google or its partners; the CWE-Bench score came from an external benchmark run by Collinear, but we found no independent replication of the results. Organizations in Fairwind's target groups, including critical infrastructure operators in healthcare, telecommunications, energy and finance, can apply for access through a Google DeepMind form, and Google says it vets all applicants; any Google Cloud customer can use CodeMender with publicly available models, according to Google. Either way, set rules before a pilot: who reviews AI-written patches, how they are tested, and which teams hold access, since Fairwind itself requires limiting use to security staff with multi-factor authentication.
Boards and CFOs should note the pace. Three Flash models in six weeks means the current benchmark leader can be replaced within a quarter. Contracts that lock in a specific model version, or skip a process for evaluating new ones, will age quickly.
The bigger picture
Google is gating its most cyber-capable model to vetted defenders first. It said early access gives defenders time to harden systems before attackers can exploit similar capabilities, and that the Cyber model ships with looser cyber safeguards than the public 3.8 Flash, which is why access is restricted. For buyers, that means the strongest AI security features may arrive through vetted programs with usage rules rather than through public APIs.
What happened next
Google kept extending the 3.8 family through September. On September 15, 2026, it launched Gemini 3.8 Live and 3.8 Live Extended Thinking, voice models available through the Gemini API, Google Workspace and the Gemini app. On September 23, it released Gemini 3.8 Flash TTS and 3.8 Flash-Lite TTS text-to-speech models, and on September 24 it added Gemini 3.8 Live with Live Avatar, a talking video persona, in Gemini Enterprise.
The next dated milestone is December 31, 2026, when the introductory 3.8 Flash price ends. Google also said Fairwind will expand partner access over time; watch for which sectors and countries it adds.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made with our newsroom’s technology tools, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is not published automatically; it is held for the editor, who decides whether it is fixed, published or dropped.
- Archive story
- It was written after the event. Its dates were checked against the date of the event, including the dates of any later developments it reports.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
