GOOGLE DEEPMIND / MODEL PROFILE
GeminiMake sense of more than words.
Gemini 3.8 Flash
An especially relevant candidate when your source material mixes documents, images, recordings, and video.
Research checked September 5, 2026
THE ASSESSMENT
Documented capabilities and limitations, with practical interpretation.
Tokens are small units used to represent inputs and outputs. API rates are usage charges for software, separate from chat subscriptions. Prices are in US dollars, as checked for this edition.
Read our research method →Where it stands out
Mixed media in one model
The API accepts text, images, video, audio, and PDFs. This is a concrete advantage for workflows such as comparing a presentation with a recording, or organizing a collection of visual and written material.
Google · Gemini 3.8 Flash specifications ↗Large context with useful tools
Google lists an input limit of 1,048,576 tokens, alongside code execution, file search, and search grounding. These are useful building blocks for a workflow that connects an answer to documents and external information.
Google · Gemini 3.8 Flash specifications ↗Lower introductory API rates
Standard pricing through December 31, 2026 is $0.75 per million input tokens and $3.75 per million output tokens, including thinking. This makes repeated trials more affordable than the premium tiers profiled here.
Google · API pricing ↗Where to be careful
Understanding media is not generating it
3.8 Flash outputs text. Its model endpoint does not generate images or audio and does not support the Live API; those experiences use other models. Computer use is still marked Preview.
Google · Gemini 3.8 Flash specifications ↗The current price is temporary
Google lists standard rates doubling on January 1, 2027. Search grounding and caching can add costs. Budget against the applicable future rate if you are building a lasting workflow.
Google · API pricing ↗Data terms depend on the service
Google’s unpaid API and AI Studio terms can permit product improvement and human review; paid-service treatment differs, with regional and account exceptions. Check the actual service, billing status, and terms before uploading confidential material.
Google · Gemini API data terms ↗OUR SUGGESTED TRIAL
Try it on your kind of work.
Use a public webinar recording and its slide deck. Ask for a short briefing with timestamps and slide references, distinguishing what the speaker said from what the slides actually show.
What to look for
Open the cited moments. Check names, amounts, speaker attribution, and anything on a visually dense slide. A large input limit does not ensure every detail receives equal attention.
An editorial test idea, not a reported benchmark result. Use public, fictional, or properly approved material.What else to compare
Google also offers Flash-Lite tiers for volume and separate image, speech, and live models. Pick by the required input and output, not by assuming every Gemini capability lives in one endpoint.
Google · Gemini release history ↗Google · Gemini 3.8 Flash specifications ↗No model in this guide replaces source checking, professional judgment, or a review of the application’s data terms.
Sources & further reading
Primary documentation checked for this edition. Provider claims are not an independent head-to-head test.