Gemini 3.8 Flash is Google's workhorse model, built for long-horizon software engineering, autonomous agents and multi-step reasoning in specialist fields. It shipped on September 2, 2026, three weeks after Gemini 3.7 Flash, and keeps the same launch rate of $0.75 per million input tokens. On DeepSWE v1.1 it now solves engineering problems end to end better than most larger frontier models. Paying Gemini app subscribers get it too, alongside AI Mode in Search and Gemini in Sheets.
Gemini 3.8 Flash takes smaller reasoning steps, calls tools iteratively and verifies its output along the way, a deliberate design choice for long-running jobs. The context window reaches one million tokens, with text, images, audio and video accepted as input and 65,536 tokens of output. You hand it a full repository, it reads, patches, runs the tests and loops until everything goes green.
That stubbornness shows up in the numbers. On DeepSWE v1.1, a long-horizon software engineering benchmark, it climbs to roughly 71% against 65.3% for version 3.7, and it posts 54.9% on HLE-Verified. Running autonomous AI agents is where it earns its keep, with Google also reporting gains on professional tests such as Vals Finance Agent V2 and Harvey's legal benchmark.
Google has been shipping a fresh Flash model roughly every three weeks since late July 2026, and Gemini 3.8 Flash is the third in that run. The pace is almost dizzying, especially with the Gemini 3.5 Pro announced at I/O in May 2026 still nowhere in sight.
Same day, second announcement. Gemini 3.8 Flash Cyber, restricted to the Fairwind Program, hunts security flaws on its own. The Chrome security team credits it with 2.6 times more correct vulnerability patches than much larger commercial models, while Wiz measured higher recall on its penetration testing benchmark at a sharply lower cost.
| Model | Release | What changed |
|---|---|---|
| Gemini 3.6 Flash | July 21, 2026 | Start of the rapid release cycle |
| Gemini 3.7 Flash | August 13, 2026 | Focus shifted to coding |
| Gemini 3.8 Flash | September 2, 2026 | Long-horizon agents, expert-level analysis |
| Gemini 3.8 Flash Cyber | September 2, 2026 | Security work, gated behind Fairwind |
Consumer access runs through the Gemini app, AI Mode in Google Search and Gemini in Google Sheets, with a Google AI Pro or Ultra subscription. Developers find it in the Gemini API, Google AI Studio, Android Studio, Antigravity and Gemini Enterprise.
Then comes the bill. The launch rate matches version 3.7 at $0.75 per million input tokens and $3.75 per million output tokens, thinking tokens included (that detail hides in a footnote of Google's own benchmark table). Both figures double on January 1, 2027, while batch mode cuts the tab in half. Before locking a budget, check Google's official rate card, since it shifts with every release.
No, not in the Gemini app, where selecting it requires a Google AI Pro or Ultra subscription. Developers pay per token through the API, at an introductory rate that runs until December 31, 2026. Google AI Studio remains the simplest place to try it before committing any budget.
Because it works harder, in Google's own words. On complex tasks it runs extra reasoning steps and calls tools repeatedly, pushing cost per task up by roughly 40% even though the per-token price is unchanged. You can dial down the effort level, or simply keep running version 3.7.
Only trusted government agencies, critical infrastructure operators and software maintainers, through the new Fairwind Program. Google is keeping tight control of a model that discovers vulnerabilities autonomously, after its cloud research team found a critical flaw in under two hours, a job that usually takes months of work.
March 2026 at the latest, with cutoff dates ranging from January 2025 to March 2026 depending on the domain. For fresher information, the model leans on external tools such as web search, billed separately through the API.
Verdict: For running coding or research agents all day long without frontier-tier invoices, this Flash has become a very strong pick, and teams already living inside Google's ecosystem will get the most from it.
Claude Fable 5.1 
Claude Fable 5 
Grok 4.6 
