Mid-Autumn Festival has just passed, and National Day is here, but the AI world hasn't taken a break. New models continue to launch one after another. Flagship models are busy pushing the boundaries of capability, while speed- and cost-oriented models are competing for users' daily work. However, as the latter's capabilities improve, the division of labor between the two is becoming blurred.
What can a model with "Flash" in its name achieve when given tasks that previously required a flagship model? #Frontend creation is an intuitive window for observation—color schemes, layout, animation, and interaction—putting works side by side reveals the differences.
Last month, Anthropic released Claude Opus 5.5. On X, we saw people using it for dynamic portfolios and ink-wash animations. These code-generated works already have their own styles. Recently, APPSO discovered that a domestic model is also becoming a new favorite among X users: the recently launched MiniMax M3.1 Flash Preview.
The anonymous model previously known as Space Bunny, which caught developers' attention, has been precisely linked to the #MiniMax M series. Even Dax, co-founder of OpenCode, suggested trying Space Bunny with the same prompts used to test Opus 5.5. Although it's unclear whether "Space Bunny" is indeed from MiniMax, the name M3.1 Flash Preview easily brings speed and cost-effectiveness to mind. But we wanted to verify whether it can compete with a flagship model like Opus 5.5 in frontend creation.
So, in MiniMax Code, we used the publicly available prompts from high-quality Opus 5.5 cases, comparing visual style, motion choreography, and overall completion, then observed its multidimensional capabilities through more topics. When the works are in front of us, whether the Flash model can produce near-flagship-level results has a more intuitive answer.
Same Prompt, Different Solutions: Design on Par
The requirement was simple: create a 15-second dynamic design portfolio that shows you are an excellent motion designer, and deliver an MP4 file. No specified colors, fonts, or storyboard. Apart from duration and purpose, the model decided everything.
In the video generated by M3.1 Flash, large off-white text fills a black background, with fluorescent green glowing from behind the letters. Text exits, rings and triangles move along orbits, followed by waveforms, jumping numbers, and a capability list. In just 15 seconds, typography, geometric shapes, and data animations appear in turn, consistently using the black-and-fluorescent-green palette.
This creative direction comes from an Opus portfolio on X. We used the author's public prompt and only added the MP4 delivery requirement. The Opus version on the left showcases typography, easing curves, and color block transformations like a series of motion exercises; the M3.1 Flash version on the right incorporates rings, waveforms, and numbers into a black-and-fluorescent-green visual system, feeling more like a tech-themed short film.
If I were to choose a direction for a tech project, I'd lean towards the M3.1 Flash version. The size and density of large text, thin lines, and charts vary, the color scheme remains consistent, and there's visual connection between segments. Opus displays more intuitive motion types, while M3.1 Flash organizes the segments more like a single film. At least for this task, M3.1 Flash Preview already possesses frontend creation capabilities comparable to Opus 5.5, and in terms of style consistency and work organization, it delivers results that better meet my needs. Choosing it doesn't require convincing yourself to accept a noticeably inferior design.
Multimodal Understanding Hidden in Scene Details
A beautiful tech short film isn't enough to show the model can handle different design intentions. Changing topics further reveals whether it will always stick to the same visual habits.
"The Meaning of Life" is another creative idea from X user @HarveenChadha. Given to M3.1 Flash, it became a hand-drawn short film. On beige paper, green hills spread out, and a small figure begins to ask what it's all for. Then the scene shifts to graph paper, with signposts reading "More," "Faster," "Perfect" appearing along the way. After a full circle, the question mark in the frame only grows larger.
The question mark gradually decomposes, and the story lands on small things. Someone prepares food, a joke amuses a friend, a planted seed grows, and there's rain, a cat, and a small artwork made by oneself. The model arranges these fragments into a wall of paper slips. Earlier, it was searching for a big answer; later, it begins observing a meal, a plant. I think the most fitting design choice here is the journal format. Paper slips, pencil lines, and small illustrations run throughout, carrying both the opening inquiry and the later recording of ordinary moments. The visual form connects with the story content, making it easier for viewers to follow the narrative.
We then tried to replicate the case of "Little Tadpole Looking for Mom" created by X user @akokoi1 with Opus 5.5. The result: on a light paper surface, the veins of lotus leaves are clearly visible. Tadpoles with thin tails swim across the pond, encountering fish, ducks, turtles, and crabs along the way, finally reaching the frog. Subtitles appear with the story, and ink wash, paper, and green lotus leaves form the film's base tone. Following the creative requirements of the Opus 5.5 case, M3.1 Flash also presented a complete ink-wash story. Rather than replicating a single frame, it better maintains the paper texture, character movement, and subtitle narrative throughout the film, collectively sustaining the style.
In another test case of the New York skyline, I chose to let a timeline control the scene changes. Initially, an orange-red sky reflects on the water, with bridges and buildings blocking part of the sunset. Dragging the bottom timeline, the sky darkens, windows light up one by one, and the river surface shifts from warm reflections to fragmented night reflections. I prefer this version's treatment of the water. During the day, large areas of warm color are visible; at night, lit windows leave fragmented light on the water. Bridges, buildings, sky, and reflections change along the same timeline, and stopping at different times yields different scenes.
Dynamic short films have the model arrange the viewing order, while interactive works give some control to the user. Wherever the timeline is dragged, the scene must present a corresponding state, with sky, lighting, and water surface staying coordinated. In games, the visuals must continuously change with player actions.
We then pushed the requirements to an arcade kart racing game, with bright blue skies, lush green grass, and red-and-white curbs forming the track, while car selection, racing, and results screens string together a complete race flow. An underwater exploration game shifts the pace: a submarine slowly navigates a submerged botanical garden, with teal fog, suspended particles, and searchlights establishing depth. The inertia retained when the submarine turns, and scanning prompts that gradually appear near landmarks, create a serene atmosphere for exploration and leave room for players to discover details. A narrative game with a little ghost hides its scene in the night of a late-autumn farm, with a bonfire illuminating nearby corn and wooden fences, and warm indoor light and green door light guiding the investigation. I like the large amount of black space it preserves; limited lighting makes small objects like the rabbit, relics, and lamps more prominent, giving each approach and interaction a sense of story.
The demands of frontend works become complex precisely in these details. Visual design, animation logic, and user operation jointly determine the experience; doing one part well alone is not enough to complete the whole work. From dynamic portfolios to hand-drawn and ink-wash styles, to interactive scenes and game prototypes, M3.1 Flash Preview and Opus 5.5 each have their trade-offs. Flash's positioning does not become a significant disadvantage in work completion, and it demonstrates the value of organizing visuals around different creative ideas and letting design serve viewing, operation, and exploration.
From Height of Intelligence to Breadth of Use
When M3.1 Flash Preview delivers results that can go head-to-head with Opus 5.5 in frontend creation, the reasons to choose Flash increase. Beyond speed and cost, the design and completion of the works themselves can also justify its use. Behind the frontend performance is a combination of design understanding, aesthetics, and coding ability. It is said that the M3 series has undergone #multimodal mixing from the pretraining stage.
[AgentUpdate Depth Analysis] This hands-on test of MiniMax M3.1 Flash Preview reveals a key trend in the AI Agent ecosystem: the "flattening" of model capabilities. Traditionally, flagship models handle complex reasoning and creation, while lightweight models manage simple tasks. But M3.1 Flash has approached—and in some dimensions surpassed—Opus 5.5 in tasks requiring multimodal understanding and code generation, such as frontend design. This means future Agent systems will no longer rely on a single "brain" but can dynamically dispatch models of different specifications based on the task, achieving an optimal balance among cost, speed, and quality. For developers, this lowers the barrier to building high-quality Agent applications and prompts a rethink of model selection strategy—no longer blindly pursuing the largest parameters but finding the "good enough" model for the scenario. At the same time, it places higher demands on Agent frameworks: how to efficiently route tasks, combine model outputs, and ensure consistency will become the next competitive focus. MiniMax's multimodal pretraining of the M3 series demonstrates the deep potential of lightweight models in vertical scenarios, which may accelerate the evolution of Agents from "general assistants" to "domain experts."
