Two nights ago in the early morning, I came across a piece of news, and it instantly perked me up.
A company ranked 9th globallyAI Lab, quietly made all three of its core models completely free.
Text, images, video—all free. No playing games with you—it's not some stripped-down version or rate-limited version. You can use it just by registering.
What kind of move is this? This style feels unfamiliar—sure enough, it's not from our country. This team from Singapore is a tough player: three models pushing forward together, and all of them have backing from international leaderboards.
Recently, it's also been highly discussed online.
In today's article, K-jie will do hands-on tests to see whether it's truly free and truly usable.
01. This new global player in the AI industry is starting to emerge
Long-time readers know that when K-jie looks at AI, there's a habit: never trust official announcements, and never get carried away no matter how flashy the demo is. I only focus on two things: leaderboard rankings and blind evaluation results.
AgnesThe three leaderboards it entered this time all happen to test "ability to get work done."
First, let's talk about the text model Agnes-2.0-Flash. The leaderboard it entered is called Claw-Eval.
The Claw-Eval leaderboard is recognized in developer circles as an Agent real-world combat medical report. Agnes-2.0-Flash surpassed both Gemini Flash and MiniMax M2.7 in some tasks. A free model being able to go head-to-head with paid flagship models on this leaderboard—you can judge the value for yourself.
Next, the image model Agnes-Image-2.0-Flash. The leaderboard it entered is called the Artificial Analysis image editing blind evaluation leaderboard.
That is, all generated images are not allowed to be labeled with which company's model produced them, and the scorers rely purely on visual judgment. Brand halo and preconceptions all fail here.
Agnes managed to squeeze into the front ranks of this leaderboard. Friends, think about it: on the image editing track, it's all veteran models that only do images. For a company that also does text and video to come over and take a front-rank spot is somewhat unfair play.
Finally, the video model Agnes-Video-2.0 entered Artificial Analysis'sImage-to-Video(with audio) leaderboard.
Image-to-video is not hard, but doing audio and audio-visual synchronization well raises the threshold. Throw in a static image, and what comes out is a dynamic short video with natural motion and ambient sound effects. For teams that need to mass-produce materials, this capability combined with a free API makes the cost-performance ratio maxed out.
Text, image, and video—three lines each made it onto leaderboards, mutually corroborating. A single subject breaking out can be luck, but all three subjects entering can only be strength. There's nothing more to say.。
02. Firsthand hands-on test
After looking at the leaderboards, K-jie's hands got itchy. Talking without practicing is fake. I directly picked some scenarios that most test the fundamentals for each of the three models, and will run through them hands-on for friends.
Text model Agnes 2.0 Flash
AI marketing overseas expansion tool website
AI overseas expansion tools are very hot right now. I made a website with 3D effects. Visually, a particle Earth with connecting lines rotates, maxing out the tech feel. And it's not a dry static webpage: interactive buttons can be clicked, pop-ups have entrance animations, numbers jump and increment, and cards have light effects on hover. The structure is also complete, from the hero screen, features, process, pricing, to CTA, guiding users step by step.
AI marketing content generator
The highlight of this case is that it turns the concept of AI-generated content into a real interactive experience. You input product information, choose category and style, click Generate, and you can see the process of the copy being generated.
The most intuitive part is those 3D flip cards—you see the product on the front, and flip it over to see the AI-written copy. This interaction is indeed very intuitive.
Agent Product Research and Analysis Report
This case turns in-depth research on 17 products into a beautifully designed interactive presentation. I went straight from searching, organizing, and analyzing to outputting a complete HTML slide deck, with no external tool dependencies throughout.
The report progresses layer by layer from market overview → product analysis → ranking comparison → pricing strategy → capability matrix → trend insights → overseas opportunities, with clear logic. Agnes 2.0 Flash's structured expression is quite impressive.
Image Model Agnes image 2.1 Flash
Miniature World × Fantasy Architecture
This case chose 'a glass castle on a dewdrop,' a creative concept that can max out the material complexity the model is capable of.
Look, everyone—at least three transparent materials have to interact simultaneously in the scene. The dewdrop is water refraction, the castle is glass light transmission and reflection, and the crystal tower has internal glow.
With these three things together, light passes into the dewdrop, then refracts onto the castle, then reflects to the surroundings—physically this is extremely complex. This is testing the depth of the model's understanding of materials in the real physical world.
Cyber Neon City Nightscape
Suitable for header images and video covers for tech-related content, with strong visual impact and the ability to show the model's power at a glance.
One image simultaneously hits three of the hardest points. Neon is multiple light sources, and the model must make each sign's light correctly land on walls, umbrella surfaces, and trench coats, while the light colors also blend with each other. If the lighting and shadows blur, everything collapses.
The toughest part is the Chinese text on the neon signs. Curved, glowing lettering is the easiest to get wrong in strokes or blur into light spots. Being able to render 'The Future Has Come' and 'City of AI' clearly and with a glowing texture means the text rendering is basically top-tier.
Amazon E-commerce Main Image
The Amazon main image is a standard test for commercial images, with hard rules and low tolerance for error; whether it performs well or not can be judged at a glance.
Prompt: pure white background, a dark gray wireless earphone charging case, lid open, product centered, commercial studio shot, high-definition details.
The purity of the white background really shows skill. Being able to press the background to true 255 pure white while keeping the product edges crisp and not blurry shows that the model has an accurate understanding of the commercial context of 'studio shooting.' The earphone case's matte shell and internal metal contacts should present different reflections. A model that can pull apart the textures stands in the first tier of material understanding.
This group passes cleanly and neatly, showing that the model can withstand the real demands of e-commerce product promotion.
Video Model Agnes Video v2.0
First generate a 5-second short video, with the camera slowly panning from right to left, the scene content moving naturally, accompanied by the ambient sound of a Tokyo city street at night.
The camera pan is smooth, with no screen tearing or dropped frames. Combined with the free API BUFF, the cost of batch production can be directly ignored.
Now look at a more cinematic shot: the overall atmosphere is very grand. An alien lands on Mars in a spaceship, the alien comes out of the spaceship, the protagonist slowly advances, yellow sand drifts in the wind all around, with a strong blockbuster feel.
The following case tests character expressiveness: a young woman stands by the window, smiles and turns her head to look at the camera, and the camera slowly pushes in. With tears in her eyes, she says, 'I'll support you.'
In the style of 1970s American film, a man and a woman dance the tango to music at a ball, while other people at the ball watch and applaud. In the scene, everyone's expressions, movements, and demeanor are all within the setting.
Finally, let's take another look at the film effect in Hayao Miyazaki's animation style: a little girl crouches in front of a stray dog and says, let me take you home. The puppy whimpers and nods.
The most brilliant part of this entire case is its coverage, from zero-cost empty-scene footage, to sci-fi blockbusters, and then from emotional performances, to retro group portraits, to animation style. It just happens to verify each of the five most valuable uses of video models.
Whether for product promotion, short dramas, popular science, or stylized content, it can hold up.
A free video model can produce this kind of effect, and its value is right there for all to see.
03. Some sharing
After talking about the rankings and running the cases, K-jie wants to say a few heartfelt words.
Over the past year or so, model capabilities have soared, and API prices have soared too. Agent has become the mainstream direction, but it is inherently a huge token consumer, with multiple steps, multiple tools, and repeated calls, every one of which burns real money. When K-jie runs complex tasks herself and uses paid models, she really has to do some mental preparation before clicking run.
If this keeps up, the only ones who can play with Agents on a large scale will forever be those few big companies with ample budgets. Small and medium-sized developers, independent creators, and friends still in school are almost being deprived even of the right to trial and error.
With Agnes opening for free this time, what K-jie sees is not marketing, but an attitude:At this AI table, it should not only be people who can afford the chips who get to sit down.
A small two-person team, after integrating Agnes, stands at the same starting line as a hundred-person big company in terms of model capabilities. The ideas in your head can immediately be run and verified, and the words 'trial and error cost' can temporarily be deleted from your dictionary.
K-jie herself has already switched her default model for daily experiments over to it. That sense of freedom from no longer having to stare at the bill and gasp for breath is really so refreshing.
The free APIs for Agnes's three models are now fully open, friends can go straight to the official website to get the freebies~
Original link:This globally ranked No. 9 AI Lab, API completely free...