AI News Feed
Market watch
Large Language Models

MiniMax M3.1 Flash Preview Matches Opus 5.5 in Front-End Creation Tests, ifanr Reports

MiniMax's newly launched M3.1 Flash Preview produced front-end design, animation and game prototypes comparable to Claude Opus 5.5 when tested with the same public prompts, according to ifanr.

The first test asked the model to make a 15-second motion design portfolio that would demonstrate the skills of an outstanding motion designer and deliver an MP4 file. No colors, fonts or storyboard were specified. According to the report, M3.1 Flash produced a black-background film with large off-white type, fluorescent green light behind the letters, rings and triangles moving along tracks, waveforms, jumping numbers and a capability list. The report said the result maintained a unified black-and-fluorescent-green visual system and organized its segments more like one film, while the Opus 5.5 version showed more types of motion more directly. The reviewer said he would lean toward the M3.1 Flash version for a technology project.

The report also tested other creative prompts. A prompt inspired by X user @HarveenChadha's idea about the meaning of life became a hand-drawn short on beige paper, with green hills, a small figure asking questions, graph paper, road signs reading "more," "faster" and "perfect," and a final sequence of small daily moments arranged as a paper wall. The report said the notebook form connected the opening question with ordinary details. In a recreation of X user @akokoi1's Opus 5.5 case "Little Tadpoles Look for Their Mother," M3.1 Flash produced an ink-wash story with visible leaf veins, tadpoles swimming past fish, ducks, turtles and crabs, subtitles and a consistent paper-and-lotus-leaf style.

Interactive tests included a New York skyline scene controlled by a timeline. Dragging the timeline changed the sky from orange-red to dark, lit windows building by building and turned warm water reflections into fragmented night reflections. The report said the bridge, buildings, sky and reflections changed along the same timeline and produced different images at different times. The model also generated an arcade kart game with a blue sky, green grass and red-and-white curbs, plus vehicle selection, racing and results screens; an underwater exploration game with a submarine moving through a flooded botanical garden, green fog, suspended particles and a searchlight; and a ghost narrative game set on a late-autumn farm at night, with a campfire, corn, wooden fences, warm indoor light and green door light guiding investigation.

According to ifanr, the M3 series used multimodal mixed training from the pretraining stage, and visual-code coordination is one way to understand its front-end ability. The report said creators can hand more specific design decisions to the model and spend their time setting direction and judging results. For many daily tasks, the report argued, users no longer need to call the strongest flagship model every time. As agents execute searches, organizing, tool calls and content generation in sequence, a single task may require dozens of model calls, accumulating waiting time and cost.

The report said the front-end examples correspond to work needs: motion portfolios can show design ability, hand-drawn and ink-wash shorts can provide creative drafts, and small games can validate campaign gameplay. Changes to colors, motion and interaction can each trigger new model calls. For MiniMax M3.1 Flash Preview, whose main scenario is coding, sustained delivery of acceptable results would let speed and cost advantages help creators try more options within a limited budget. The report described speed, quality and price as an impossible triangle, but said Flash models are changing the exchange ratio among the three, allowing many coding tasks once handled only by flagship models to be completed faster and at lower cost.

The report compared Flash models with flagship models to PC compatibles and mainframes. Mainframe-era computing concentrated expensive capacity in a few institutions, while PC compatibles used standardization and scale production to bring computing into offices and homes. If intelligence becomes a standard component that can be called at any time, a larger application market can emerge. After PCs became widespread, industrial value shifted toward operating systems, chips and software ecosystems; after Flash models become widespread, AI value may flow more toward agents, workflows, proprietary data and application entry points. Flagship models continue to explore the boundaries of intelligence, while models such as MiniMax M3.1 Flash Preview bring proven capabilities into high-frequency development. The report said future profits may belong to products that know how to use that intelligence.

Editor's Summary MiniMax's M3.1 Flash Preview, a new speed- and cost-oriented model, was tested by ifanr against Claude Opus 5.5 on front-end creative prompts and generated motion, hand-drawn, ink-wash, interactive and game prototypes, with the reviewer preferring it in some style and organization aspects. The report argues that Flash-class models can compete with flagships in daily coding and design workflows, shifting AI value toward agents, workflows, proprietary data and application entry points. These claims come from a single publication's tests and do not establish independent benchmark results.