AI Daily: Google Launches Gemini 3 Flash for Speed and Cost Efficiency, OpenAI Opens ChatGPT App Store

In this wave of AI, December seems to be the key moment for tech giants to flex their muscles. Google not only updated its models but also pushed the battle to the extreme balance of “speed” and “utility”; OpenAI chose to expand its ecosystem, allowing developers to truly build business models on the ChatGPT platform; while Microsoft quietly dropped a bombshell in the 3D generation field.

This article will take a deep dive into these three major updates to see how they impact our work and creativity.

Google Gemini 3 Flash: The Sweet Spot of Speed and Intelligence

If you’ve been following AI model development, you might have noticed a phenomenon: usually, we have to choose between “smart but expensive and slow” and “fast but slightly dumber” models. But Google’s release of Gemini 3 Flash seems to break this established rule.

This new model is the latest member of the Gemini 3 series, and its core selling point is very clear: extreme speed and cutting-edge intelligence, at a very low cost.

According to Google’s data, Gemini 3 Flash’s reasoning capabilities have surpassed the previous Gemini 2.5 Pro, yet it is three times faster. This is a staggering figure, meaning developers and enterprises no longer need to sacrifice model intelligence to save money or pursue instant responses. It scored an impressive 90.4% on GPQA Diamond (graduate-level Q&A test) and performs on par with its big brother Gemini 3 Pro even in multimodal understanding (processing video and images).

Why Does This Matter to General Users?

Starting today, free users worldwide can access this model in the Gemini App. You might feel that response speeds are faster, and the ability to handle complex instructions has improved. Especially when dealing with video understanding or long document analysis, this low-latency experience makes AI feel more like an on-call assistant rather than a server still spinning its wheels.

A Boon for Developers: Major Update to Gemini CLI

For engineers who stare at the terminal all day, good news has arrived. Google simultaneously announced that Gemini 3 Flash is officially available in the Gemini CLI.

Why is this important? Because working in a terminal environment demands high frequency and fast pace. If you have to wait ages for AI to think after every command, that “flow” state is broken. The addition of Gemini 3 Flash allows developers to perform so-called “Agentic Coding” directly in the command line.

For an interesting example, Google demonstrated using Gemini 3 Pro in the CLI to generate a 3D Voxel simulation of the Golden Gate Bridge in San Francisco. This task used to require expensive models, but now Gemini 3 Flash can handle such tasks requiring high logic and creativity at less than a quarter of the price of the Pro version.

Furthermore, for backend engineers who need to conduct stress testing, Gemini 3 Flash can quickly generate and fix Python scripts to simulate various user scenarios (such as payment failures, inventory timeouts), greatly saving time on hand-writing test scripts.

OpenAI App Submission Open: ChatGPT’s App Store Moment?

Turning the lens to OpenAI, they are doing something that could completely revolutionize the AI business model. Starting from December 17th, developers can submit applications to ChatGPT.

Does this sound a bit like the previous “GPTs”? It’s actually quite different.

This time, OpenAI released a brand new Apps SDK (currently in beta). This allows developers to build “Chat-native” experiences. These apps are no longer just simple text conversations; they can execute specific actions directly within the chat window. Imagine discussing dinner plans with ChatGPT and then directly placing a grocery order through an App, or calling an App to generate and present slides while discussing a presentation structure.

Key Differences and Commercialization

What excites developers most is undoubtedly the possibility of commercialization. In the initial phase, OpenAI allows developers to guide users to their own websites or native Apps to complete transactions for “physical goods”. Although currently focused on physical goods, they also revealed that they will explore monetization models for digital goods in the future.

OpenAI also simultaneously launched the App Directory, which is like an App Store inside ChatGPT. Users can browse and search for various applications within it. Once connected, these Apps can be summoned at any time via @ mentions or through the tool menu.

This represents ChatGPT transitioning from a “chatbot” to an “operating system”. For developers, this is an opportunity to seize the initiative; for users, the future ChatGPT will become more versatile and proactive.

Microsoft TRELLIS.2: The Magic of Generating High-Quality 3D Models from Single Images

Although Google and OpenAI have occupied most of the headlines, Microsoft’s progress in the field of graphics should definitely not be ignored. They released TRELLIS.2, a powerful model with 4 billion parameters, specifically designed to solve a long-standing pain point: how to generate high-quality 3D assets from a single 2D image.

Technical Highlights

TRELLIS.2 adopts a technology called Flow-Matching Transformers, combined with a Sparse Voxel 3D VAE architecture. Without getting into obscure jargon, simply put, it achieves the following:

1. High-Resolution Geometry and Texture: The 3D models it generates are no longer a blurry mess but have fine geometric structures and PBR (Physically Based Rendering) textures.
2. Amazing Efficiency: Through 16x spatial compression technology, it remains efficient when generating complex objects.
3. Diverse Outputs: Whether it’s hard-surface mechanical structures, organic character models, or translucent objects, it handles them quite well.

For game developers, 3D artists, or metaverse builders, this is a tool of great practical value. You can try it out yourself on the Demo page on Hugging Face, or download the model directly for local deployment. This will significantly lower the barrier to 3D content creation.

Frequently Asked Questions (FAQ)

Q1: Is Gemini 3 Flash free? Yes, for general users, Google has already started pushing Gemini 3 Flash to the Gemini App as a free default model. For developers, it currently offers preview pricing in the API and CLI, which is very low compared to the Pro version.

Q2: Can I sell things on ChatGPT now? OpenAI currently allows developers to submit apps and link to external websites for “physical goods” transactions. As for digital goods sales and more complete payment integration, OpenAI stated they will gradually explore and release more information in the future.

Q3: Does the Gemini CLI update affect non-programmers? The main impact lies in the stability of backend services and development speed. Although non-programmers won’t use the CLI directly, if the apps you use are built by developers using Gemini CLI, update speeds might increase, and features will be more stable.

Q4: Can 3D models generated by TRELLIS.2 be used in commercial games? TRELLIS.2 is an open-source project, and according to its release information, it can generate high-quality assets. However, for specific commercial license terms, it is recommended to refer to the detailed instructions on its GitHub page, especially regarding the copyright ownership of generated content.

Q5: Why does Google emphasize the “multimodal” capabilities of Gemini 3 Flash? Because in real-world applications, we process not just text. Being able to accurately understand video, audio, and images simultaneously and respond quickly is the key for an AI assistant to transform from “fun” to “useful”. For example, you can let it watch a golf swing video and immediately get improvement suggestions; this is the power of multimodal combined with speed.