AI Competition Intensifies: OpenAI Garlic Takes on Gemini 3, Alibaba Z-Image's New Model Challenges Flux Head-On

The competition in the current AI industry has entered a phase of "close combat", with fierce confrontations breaking out simultaneously in two major battlefields: general-purpose large models and professional image models. On one side, OpenAI has urgently unveiled its new weapon codenamed Garlic to face the impact of Google's Gemini 3; on the other side, Alibaba's Z-Image series continues to make efforts, launching a strong challenge to Flux, a rising star in the image field, with a new ControlNet model. The industry pattern is being rapidly rewritten.

I. General AI Battlefield: OpenAI Garlic Debuts, a Counterattack in Response to the "Red Alert"

OpenAI's recent announcement that "the model codenamed Garlic has completed pre-training" is not an isolated technological iteration, but a precise response to the industry's competitive landscape — all of which dates back to the "red alert" issued internally by OpenAI some time ago. With the sudden rise of Google's Gemini 3, it has not only surpassed in multiple industry benchmark tests, but also driven its monthly active users to soar from 450 million to 650 million. Even corporate leaders such as Salesforce's CEO have publicly switched to its camp, directly triggering user attrition and strategic panic at OpenAI.

To address the crisis, OpenAI has activated the highest level of emergency status, suspending all non-core projects such as advertising business and personal assistant Pulse, and diverting all resources to the upgrade of ChatGPT and the development of new models. Garlic is precisely the key piece in this "defensive counterattack". According to industry insiders, Garlic's core breakthrough lies in solving the pre-training defects of the previous model "Shallotpeat", enabling it to inject a knowledge volume comparable to that of large models into a small model architecture. This means that while controlling R&D costs, it can achieve more efficient reasoning capabilities — and reasoning performance is precisely one of Gemini 3's core advantages.

From a strategic perspective, Garlic is not only a direct response to Gemini 3, but also marks OpenAI's shift in competitive strategy from "scale expansion" to "efficiency optimization". Combined with the news that it plans to release an inference model "internally evaluated to be ahead of Gemini 3" in the near future, the completion of Garlic's pre-training has undoubtedly injected important confidence into OpenAI's efforts to stabilize its user base of 800 million weekly active users.

II. Image AI Battlefield: Alibaba Z-Image Strikes Again, New Model's Precise Control Challenges Flux

While the general AI battlefield is in full swing, the competition in the image generation field is equally intense. As Flux has become a new benchmark in the field with its innovative architecture, Alibaba's Z-Image series continues to make breakthroughs as a "dark horse". The newly released model "Z-Image-Turbo-Fun-Controlnet-Union" has even been evaluated by the industry as a powerful work that "outperforms Flux".

The core competitiveness of this new model lies in its enhanced ControlNet capability — the R&D team has integrated the ControlNet structure into 6 key blocks of the model, enabling it to accurately respond to various image control conditions. It can achieve millimeter-level precise control over everything from human poses and movements to the edge contours and spatial depth of objects. This technological advantage is directly translated into clear scenario value: in the scenario of human pose generation, designers can obtain the desired action shapes without repeatedly adjusting prompts; in the field of architectural design, the model can quickly render detailed and proportionally accurate design drawings by simply inputting a simple line drawing.

More importantly, the model is highly compatible with mainstream workflow tools such as ComfyUI, and can seamlessly integrate into the production chain of professional creators, greatly improving the efficiency of links such as character pose design and architectural rendering. As the Z-Image series continues to evolve from basic image generation to "precise control", it has not only gained a firm foothold in consumer scenarios, but also quickly narrowed the gap with leading models in the professional field. As predicted by the industry, if this iterative speed is maintained, Z-Image is expected to become a qualified competitor to Nano Banana (Google Gemini's ecological image tool). Developers and designers interested in this powerful ControlNet model can directly visit z-image.me to experience its precise control effect in actual workflows.

III. New Logic of AI Competition: From "Parameter Competition" to "Scenario-Focused Breakthrough"

Whether it is OpenAI Garlic's "efficiency first" or Alibaba Z-Image's new model's "precise control", it indicates that the competition in the AI industry has bid farewell to the era of simple "parameter accumulation". In the field of general AI, enterprises pay more attention to the response speed, personalized experience and cost control of models in actual scenarios; in the field of image AI, "generation quality" has become a basic requirement, and "controllability" and "workflow compatibility" have become new competitive focuses.

For the industry, this shift in competition is undoubtedly a positive signal — enterprises are devoting more energy to solving the actual pain points of users, and ultimately promoting AI technology from the laboratory to wider industrial applications. The showdowns between OpenAI and Google, as well as between Alibaba and Flux, will continue to bring technological breakthroughs and innovative inspirations to the industry, and the "hundred flowers blooming" of the AI ecosystem is accelerating.