Gemini 3.8 Flash: Pricing, Benchmarks & 1M Context (2026)

Gemini 3.8 Flash is the model Google (Google DeepMind) announced on September 2, 2026 — the third Flash-series release in six weeks. Pricing is $0.75 per 1M input tokens and $3.75 per 1M output tokens, unchanged from Gemini 3.7 Flash. That said, this is still an introductory price good only through December 31, 2026, after which it doubles to $1.50/$7.50. Alongside it, Google also announced a security-focused sister model, Gemini 3.8 Flash Cyber, built for vulnerability detection and automated patching.

The context window remains 1M input tokens with a maximum output of roughly 65,000 tokens (about 1,600 pages), and the model handles multimodal input — text, images, audio, video, and PDFs. Compared with Gemini 3.6 Flash and 3.7 Flash, the main changes are clear gains on coding and security-related benchmarks, plus the new security-specialized Cyber sister model. The knowledge cutoff has not been disclosed.

Pricing runs in the same two tiers as 3.7 Flash — introductory and standard. Approximate yen figures below assume 1 USD ≈ 150 JPY.

The headline price held steady versus 3.7 Flash, but the doubling on December 31, 2026 is unchanged. Any long-term cost planning should be built around the standard 2027 price, not the current introductory rate.

The official announcement reports results across reasoning, coding, and security domains.

The 54.9% HLE-Verified score reflects multi-step reasoning across STEM and specialist domains, while the DeepSWE and CyberGym results indicate the model matches or beats larger existing models on autonomous software engineering and vulnerability-discovery tasks. Some underlying scores are not disclosed beyond the figures Google published, and this article sticks to what was officially stated.

Gemini 3.8 Flash Cyber is a security-only sister model specialized for vulnerability detection and automated patch generation. Chrome Security reported that, on Chrome vulnerabilities, it "generated 2.6x more correct patches than the best commercial model." Cyber is not generally available, however — it's offered through the Fairwind Program, an application-based framework that gives priority access to trusted government agencies, critical-infrastructure operators, and software maintainers. That's a different distribution model from the regular Flash line, which anyone can access with an API key.

Gemini 3.8 Flash is available through Google AI Studio, Android Studio, the Gemini API, Gemini Enterprise, Google Sheets, and the Gemini app. It continues the rollout pattern seen since Gemini 3.5 Flash / Omni, with the Flash line steadily broadening its distribution channels. Cyber, as noted above, is limited to Fairwind Program access and isn't part of the general-availability channels.

- Coding agents: the DeepSWE v1.1 and CWE-Bench patch-generation results suggest it can compete with leading models at lower cost, making it a candidate for agentic code generation and fix tasks
- Long-context workloads: the 1M-input / ~65,000-output token budget is unchanged, so it handles large specs or whole codebases the same way 3.7 Flash did
- Multimodal input: beyond text, it handles images, audio, video, and PDFs — useful for cross-document review or meeting-note processing
- Cost planning caveat: today's $0.75/$3.75 is an introductory price good only through December 31, 2026. Any cost estimate that spans the new year should be built on the $1.50/$7.50 standard rate — otherwise the actual bill can end up double what you planned for