Google is expanding Gemini with Google Pics, a new image tool, as well as more capable models and targeted video analysis. The changes mainly affect Google Workspace users who produce presentations, documents, marketing materials, or work with large amounts of video. They are intended to shorten routine workflows, but complex tasks may require more computing work and therefore incur higher token costs.
What Google Pics brings to Workspace
Google Pics is an Artificial Intelligence (AI) tool for generating and editing images. Instead of constructing an entire graphic manually, you describe it with a prompt, meaning a text instruction that specifies what you want. You can then select individual objects, text, or areas of the image and describe further changes.
According to the Google Workspace announcement, Pics is designed to work as a standalone app and inside Docs, Slides, and eventually Drive. You could edit an image in a document or presentation without first transferring it into separate design software. Collaborative editing is also part of the product.
The everyday examples are fairly specific: you could make an event poster, generate several versions of a social media post, or adapt an illustration for a customer presentation. Google also identifies marketing campaigns and product storytelling as possible uses. Pics can produce several versions of a requested image so that you can choose the best result.
Its editing features include isolating and transforming individual objects, as well as modifying or translating text that already appears inside an image. According to The Verge’s description of Pics, the tool can also upscale images to 2K or 4K resolution and crop them for the web, social media, print, or other digital formats. Whether the output consistently meets Google’s claim of professional-grade quality has not been independently verified.
The reports do not completely agree on availability. Google says Pics can be accessed through pics.new and used in Docs and Slides. TechCrunch describes a rollout over several weeks for most Workspace customers and subscribers to Google AI Pro or Ultra. Access may therefore depend on your account and the stage of the phased rollout; the sources do not state a separate price for Pics.
What Gemini changes for demanding tasks
Alongside Pics, Google has introduced Gemini 3.8 Flash. Flash is the model family aimed at speed and a lower price-to-performance ratio. The new version is supposed to perform more reasoning steps on complex tasks and call tools repeatedly instead of producing a result after a single pass.
That approach can be useful for demanding analysis or a job with several connected stages. A model might check information, revise an intermediate result, and then prepare a final assessment. For a short piece of writing or a basic summary, that additional work is not automatically necessary.
Google released 3.8 Flash only a few weeks after 3.7 Flash. The Decoder describes it as the third Flash release in six weeks. On Google’s DeepSWE v1.1 benchmark, a standardized test collection for long-running software tasks, the model scored 73.7 percent according to the provider, compared with 65.3 percent for its predecessor. Such scores indicate a direction of travel, but they reveal less about its usefulness for presentations, research, or everyday office work.
The introductory price is reported as $0.75 per million input tokens and $3.75 per million output tokens. Tokens are small units of text that can be used to calculate AI model charges. According to The Decoder, regular prices are scheduled to increase in January 2027 to $1.50 and $7.50 respectively.
An unchanged price per token does not mean an unchanged price per task. The Verge cites Google’s warning that 3.8 Flash may use more tokens, particularly at higher effort levels. An early Artificial Analysis measurement estimated that each task cost around 40 percent more than with 3.7 Flash despite identical introductory rates, partly because of 30 percent more output tokens and additional turns. This early finding does not guarantee the same outcome for your own tasks.
What makes video analysis more efficient
For video, Google is simultaneously taking the opposite approach: processing less material by default and searching more selectively. The new agentic video analysis feature, in which the model independently chooses suitable analysis tools and video segments, is available for Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. The system decides whether a task requires video frames, audio, or the transcript.
Gemini previously sampled video at a default rate of one frame per second while also analyzing the audio track. That method can miss very brief events and consume many tokens when applied to long recordings. The new approach can revisit suspicious or relevant time windows at a higher frame rate.
One practical example is finding a scene that lasts less than a second inside hours of footage. According to Google, the system can also count repeated movements or identify unusual events by inspecting notable sections more closely. That is more useful for media archives, quality checks, or reviewing long recordings than treating every frame with equal attention.
Google DeepMind claims token savings of up to 88 percent, cost reductions of up to 66 percent, and quality gains of up to 7 percent. These are provider claims and maximum figures that have not been independently verified. The Decoder reports that the feature is initially available through the Gemini application programming interface, the technical connection used by other software; later integration into the Gemini app and YouTube has been announced.
Pros and Cons of Google’s expansion
Pros:
- Fewer app changes – You can create, edit, and review images with others within the Docs and Slides environment.
- More targeted corrections – Individual objects and text can be selected instead of regenerating the entire image after every prompt.
- More power for complex work – Gemini 3.8 Flash is intended to process multistage tasks more thoroughly than its predecessor.
- More efficient video review – Agentic analysis focuses its effort on relevant scenes and may reduce token use and costs.
Cons:
- Variable task costs – More reasoning steps and longer outputs can make a task more expensive even when the price per token remains unchanged.
- Limited comparisons – Benchmarks and provider figures say little about the reliability of your particular document, image, or video.
- Unclear creative provenance – TechCrunch notes that Pics uses an image model trained on artists’ work without offering the kind of royalty marketplace found on conventional creative platforms.
- Phased access – Not every feature is immediately available to every Workspace account, and video analysis is not initially described as a standard Workspace tool.
What this means for you
If you are a beginner, a small image project is the most sensible first step. For example, you could select an existing illustration in Slides, remove an unwanted object, and translate its embedded text for a second language version. Compare several variants and check the typography, details, and overall message before using the result externally.
If you already have experience with AI tools, it helps to separate tasks by the amount of effort they actually require. Basic drafts and short summaries do not automatically need the model that performs the most reasoning. For a substantial analysis, however, you could test whether 3.8 Flash produces a better result through intermediate steps while comparing its token consumption and output length with a simpler option.
For Switzerland, translating text directly inside images could be particularly useful for presentations in German, French, or Italian. The sources do not identify separate Swiss availability or provide evidence about the quality of individual national languages. They also offer insufficient detail about Pics data protection, storage locations, or administrative controls, so organizations must rely on the information available for their particular Workspace account.
Google’s expansion combines two different goals: bringing creative work closer to everyday office applications and directing the computing effort of demanding models and video analysis more selectively. Pics looks practical for quickly producing work graphics, but it does not remove the need for design judgment or careful review. The unresolved questions are how reliably these tools perform outside demonstrations and how strongly additional token use affects real-world costs.
Sources
- Try Google Pics: Easy image creation and editing in Google Workspace – Google, 2026-09-01
- Google’s answer to Canva is an AI tool where you prompt instead of design – TechCrunch, 2026-09-01
- Google Pics is like Canva, but with even more AI – The Verge, 2026-09-01
- Google says its new Gemini 3.8 Flash model ‘works harder’ but might cost more – The Verge, 2026-09-02
- Googles Gemini Flash 3.8 arbeitet härter als der gerade mal drei Wochen alte Vorgänger 3.7 Flash – The Decoder, 2026-09-02
- Google Gemini analysiert Videos jetzt agentisch und spart bis zu 88 Prozent Token – The Decoder, 2026-09-02
- Introducing agentic video understanding with Gemini – Google DeepMind, 2026-09-01
