Gemini 3.8 Flash Arrives in Gemini App and AI Mode

What is Gemini 3.8 Flash?

Gemini 3.8 Flash is Google’s newest Flash model, designed for long-horizon software engineering, autonomous agents, and complex workflows. Google introduced the model on September 2, 2026, and describes Gemini 3.8 Flash as its most intelligent Flash model to date. That is a vendor characterization, but the intended use cases show that Google is positioning the model for tasks that require multiple connected steps rather than a single short answer. Google’s launch announcement sets out the model’s availability and stated goals.

Gemini 3.8 Flash matters because Flash models generally target a balance between capability and cost for everyday AI tasks. A model built for complex workflows can be more useful when a request involves planning, revising, analyzing several inputs, or carrying work across a longer sequence. The practical limitation is that a model’s stated focus does not guarantee correct results for every task, especially when the work depends on incomplete, inaccurate, or sensitive source material.

Gemini 3.8 Flash users should treat the new model as an option for tasks that need more sustained reasoning, not as a replacement for reviewing important output. Software code, business documents, financial calculations, and health-related summaries still need human verification before anyone relies on them.

Where can users access Gemini 3.8 Flash?

Gemini 3.8 Flash is available to Google AI Pro and Ultra subscribers in the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets. Google announced those consumer and productivity entry points on September 2, 2026. The rollout places the same named model in a standalone chatbot, a Search feature, and spreadsheet workflows, which gives subscribers several ways to use it without working directly with an API.

Gemini 3.8 Flash in the Gemini app is likely to be the simplest route for people who want conversational help with writing, planning, analysis, or multi-step prompts. Google has continued to place specialized AI capabilities inside its consumer products, including AI music generation in Gemini. Access still depends on the user having an eligible Google AI Pro or Ultra subscription, so free Gemini users should not assume the model appears in their account.

Gemini 3.8 Flash in AI Mode gives Google another model option inside Search. AI Mode is intended for queries where a conversational response or a synthesized explanation is more useful than a conventional list of links. Users should still open and assess cited sources when an answer affects spending, health, legal obligations, or other consequential decisions.

What changes in Gemini 3.8 Flash for Google Search and Sheets?

Gemini 3.8 Flash expands Google’s AI model availability across Search and Sheets, but Google has not described a separate feature list for each product in its announcement. AI Mode in Google Search can use the model for conversational search interactions, while Gemini in Google Sheets can make the model available in spreadsheet-based workflows. The model’s stated focus on complex workflows suggests that Google expects it to handle tasks that involve several pieces of context or several connected instructions.

Gemini 3.8 Flash in Search matters because AI Mode can affect how users approach research. Google has also added AI-generated forecasts to Search through WeatherNext 3 results, showing how AI features increasingly appear alongside information people may use to make everyday decisions. AI summaries can save time, but an answer remains only as reliable as its underlying information and the model’s interpretation of it.

Gemini 3.8 Flash in Sheets may be more useful for organizing information, explaining formulas, or helping structure a spreadsheet task than for making final decisions from business data. The sensible approach is to validate formulas, source data, and calculations independently before sharing a spreadsheet with customers, coworkers, or financial stakeholders.

How do Gemini 3.8 Flash thinking-effort settings work?

Gemini 3.8 Flash offers Low, Medium, and High thinking-effort settings, and Google sets Medium as the default. Google says High thinking effort is intended for deep reasoning, mathematics, and difficult multi-step tasks. The setting gives developers and product users a way to align the model’s reasoning effort with the complexity of the request instead of applying the same approach to every prompt.

Gemini 3.8 Flash’s Low setting is the sensible starting point for simpler tasks where speed matters more than extended analysis. Gemini 3.8 Flash’s Medium setting is the default for general work, while the High setting is intended for tasks that require more deliberate reasoning across multiple steps. Google’s latest-model documentation lists the model and its supported developer access.

Gemini 3.8 Flash thinking effort does not make an answer automatically accurate. Higher effort can be appropriate for a harder task, but it cannot correct a vague prompt, false input, missing context, or an unsupported assumption. Users should provide clear source material and check the final output against authoritative records.

How capable is Gemini 3.8 Flash on Google’s benchmark?

Gemini 3.8 Flash scored 54.9% on HLE-Verified, according to Google’s model card. Google cites HLE-Verified as a benchmark for multi-step reasoning across STEM, humanities, and professional fields. The result is self-reported, so it indicates how Google evaluated the model under that benchmark rather than proving that the model will perform at the same level in every real-world workflow.

Gemini 3.8 Flash’s HLE-Verified result matters because the benchmark is intended to test difficult questions that require more than retrieval of a simple fact. A score on a demanding benchmark can help developers compare models during early evaluation, but benchmark performance does not measure every practical concern. Reliability, response consistency, tool use, data handling, latency, and domain-specific accuracy can all affect whether a model fits a particular job.

Google DeepMind’s Gemini 3.8 Flash model card provides the company’s benchmark details and model information. Developers should test Gemini 3.8 Flash against their own representative prompts before using it for production software, customer support, compliance work, or other high-impact tasks.

What does Gemini 3.8 Flash API pricing cost?

Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens at introductory paid API rates through December 31, 2026. Google says the standard paid API price rises on January 1, 2027, to $1.50 per million input tokens and $7.50 per million output tokens. The introductory rate is therefore half of the scheduled standard rate for both input and output tokens.

Gemini 3.8 Flash pricing matters most for developers building products that send large amounts of text, code, documents, or conversation history to the model. Input tokens measure material sent to the model, while output tokens measure the generated response. Long prompts and detailed responses can increase total usage even when an individual request appears inexpensive. Google also warns that complex tasks can consume more tokens because Gemini 3.8 Flash may perform additional reasoning steps and iterative tool calls, so the lowest listed token rate does not guarantee the lowest total cost for every request.

Gemini 3.8 Flash launched with no announced shutdown date in Google’s deprecation schedule. That offers a clearer starting point for teams considering the model, but developers should still plan for model and price changes because API services evolve over time. Teams that use multiple models to manage cost may also be interested in how model mixing for coding tools is becoming a practical product strategy.

Who can use Gemini 3.8 Flash in developer tools?

Gemini 3.8 Flash is available to developers through Google AI Studio, the Gemini API, Android Studio, Google Antigravity, Stitch, and Gemini Enterprise. Google announced the developer rollout alongside the consumer availability on September 2, 2026. The broad tool list means developers can access the model for direct API work, Android development, enterprise use, and Google’s other AI-oriented creation environments.

Gemini 3.8 Flash accepts 5 input types: text, images, video, audio, and PDFs. Google’s Gemini API documentation lists the stable model ID as gemini-3.8-flash, with an input limit of 1,048,576 tokens and an output limit of 65,536 tokens. The documentation also lists support for code execution, file search, function calling, Google Maps grounding, Search grounding, URL context, and preview computer use. Grounding connects a model response to an external information source or service, but each tool must be enabled and supported in the product where you are working.

Gemini 3.8 Flash is particularly relevant for developers who need a model for software engineering or agent-style workflows. An autonomous agent is software that can take a sequence of actions toward a goal, often using tools or services along the way. That capability can reduce repetitive work, but it also increases the importance of permissions, testing, logging, and human review when an agent can affect code, data, accounts, or business processes.

Gemini 3.8 Flash developers should begin with narrow permissions and low-risk test data. Developers should also keep credentials, customer records, proprietary code, and regulated information out of prompts unless their organization has reviewed the applicable Google product terms and data controls. Stop and consult an organization’s security, privacy, or compliance team before connecting an AI agent to production systems that can alter records, initiate payments, or access sensitive customer information.

What is Gemini 3.8 Flash Cyber?

Gemini 3.8 Flash Cyber is a separate, more permissive security variant that Google restricts to trusted defenders through its Fairwind Program. Google launched Gemini 3.8 Flash Cyber alongside the general Gemini 3.8 Flash model on September 2, 2026. Google limits Fairwind Program access to trusted government authorities, critical-infrastructure operators, and software maintainers. The restricted access reflects the fact that cybersecurity models can be useful for legitimate defense work while also creating misuse risks if made broadly available without safeguards.

Gemini 3.8 Flash Cyber is not the same product as the version available in the Gemini app, AI Mode, Google Sheets, and the general developer tools listed by Google. The Fairwind restriction means ordinary Gemini users should not expect access to the cyber variant through a standard subscription. Security teams that need AI assistance should use approved tools and follow their organization’s incident-response, access-control, and disclosure procedures.

Gemini 3.8 Flash Cyber also arrives as AI systems become part of both defensive and criminal activity. Reports of AI-assisted fake dating profiles show why model providers and users need to treat automated content and workflow tools as potential abuse vectors as well as productivity tools. Users who believe an account or device has been compromised should preserve evidence, change passwords from a safe device, and contact the affected service or a qualified security professional rather than experimenting with unfamiliar security tools.

FAQ

Is Gemini 3.8 Flash available in the Gemini app?

Gemini 3.8 Flash is available in the Gemini app for Google AI Pro and Ultra subscribers. Google announced that access on September 2, 2026, so users need an eligible subscription before expecting the model in the app.

Can free Gemini users access Gemini 3.8 Flash?

Gemini 3.8 Flash access is not listed for free Gemini users in Google’s announcement. Google specifically named Google AI Pro and Ultra subscribers for Gemini app access, so free users should check their account before assuming the model is available.

What is the default thinking-effort setting for Gemini 3.8 Flash?

Gemini 3.8 Flash uses Medium thinking effort by default. Google says High thinking effort is intended for deep reasoning, math, and difficult multi-step tasks, while Low effort better suits simpler requests.

How much does the Gemini 3.8 Flash API cost?

Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. Google says the standard paid rates rise to $1.50 for input tokens and $7.50 for output tokens on January 1, 2027.

Is Gemini 3.8 Flash Cyber available to everyone?

Gemini 3.8 Flash Cyber is not available to everyone. Google restricts the more permissive security variant to trusted defenders in its Fairwind Program, separating it from the general Gemini 3.8 Flash rollout.