Загрузка видео...

Не удалось загрузить видео

На главную

Kimi K3.1 Pro is among the leading models in agentic knowledge work and coding, but early signs suggest it will rank lower on AA-3D We are still collecting the preference votes for its Elo on AA-3D, our benchmark for how well AI models build interactive 3D scenes. In agentic...

ArtificialAnlys's profile picture

Artificial Analysis

165,273 subscribers

32,782 просмотров • 7 дней назад •via X (Twitter)

Наука и технологииИгры
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 9

Фото профиля Chahat Sharma
Chahat Sharma6 дней назад

how many AA-3D votes do you usually need before the Elo settles? curious how early is early here

Фото профиля Fernando Ruiz
Fernando Ruiz6 дней назад

top tier on agentic coding but weaker at 3d scenes says spatial stuff is its own skill. one leaderboard number hides a lot

Фото профиля Qiwei
Qiwei6 дней назад

I'd try it on coding agents first, 3D can wait

Фото профиля Walter Morales
Walter Morales6 дней назад

tops terminal-bench and still builds a bike trail i would not survive

Фото профиля The SI Maximalist
The SI Maximalist6 дней назад

Coding success and human preference answer different questions. I’d watch whether the AA-3D gap holds as votes accumulate; a provisional Elo rank needs an uncertainty band

Фото профиля TslaMan
TslaMan6 дней назад

It's a pretty much right now we're in the PlayStation 1 era of AI ala Final Fantasy 7

Фото профиля Owen Gidusko
Owen Gidusko6 дней назад

how many tokens and how long did it take to achieve the output?

Фото профиля 仿生狮子
仿生狮子6 дней назад

众所周知（并不）Web 能力上： Fable 5 约等于 Opus 5，而 Opus 约等于 Kimi K3.1 Pro.... 所以 Kimi 这三个月里，前端水平没有很多长进 😅

Фото профиля Finn Klein
Finn Klein6 дней назад

Its just benchmaxed

Похожие видео

Kimi K3.1 Pro from Kimi.ai is now available on Sunra, alongside K3.1 Pro Highspeed, the same model at two to three times the speed. From a sketch to a 3D model. From a photo of your room to furniture that fits it. From one prompt to a complete launch: poster, photography, product film and 3D viewer. Start your next project with Kimi K3.1 Pro at
0:32

Kimi K3.1 Pro from Kimi.ai is now available on Sunra, alongside K3.1 Pro Highspeed, the same model at two to three times the speed. From a sketch to a 3D model. From a photo of your room to furniture that fits it. From one prompt to a complete launch: poster, photography, product film and 3D viewer. Start your next project with Kimi K3.1 Pro at

Sunra

77,523 просмотров • 4 дней назад

MIXAR JUST PUT 400+ AI MODELS INSIDE ONE 3D WORKSPACE Mixar gives you access to LLMs, image-to-3D, texturing models and native 3D tools from the same environment: → One workspace → 400+ models → Agents can work directly inside the scene → Multiple agents can work across different scenes → Built on a fork of Blender, not another plugin No need to stitch together separate AI tools, scripts and 3D software. Try it now.
0:30

MIXAR JUST PUT 400+ AI MODELS INSIDE ONE 3D WORKSPACE Mixar gives you access to LLMs, image-to-3D, texturing models and native 3D tools from the same environment: → One workspace → 400+ models → Agents can work directly inside the scene → Multiple agents can work across different scenes → Built on a fork of Blender, not another plugin No need to stitch together separate AI tools, scripts and 3D software. Try it now.

marcus

19,367 просмотров • 28 дней назад

Kimi K3 vs. MiniMax M3.1 Flash | WebGPU & 3D Koi Pond Simulation One technical challenge. Two AI models. Different approaches to interactive 3D development. Testing both models on a WebGPU-powered 3D koi pond, featuring realistic water ripples, swimming koi fish, dynamic reflections, and real-time rendering. MiniMax M3.1 Flash stands out for its speed, coding capabilities, and potential to accelerate creative development workflows. The real benchmark? Turning complex ideas into functional, interactive experiences. #M31Flash #WebGPU #AI
0:23

Kimi K3 vs. MiniMax M3.1 Flash | WebGPU & 3D Koi Pond Simulation One technical challenge. Two AI models. Different approaches to interactive 3D development. Testing both models on a WebGPU-powered 3D koi pond, featuring realistic water ripples, swimming koi fish, dynamic reflections, and real-time rendering. MiniMax M3.1 Flash stands out for its speed, coding capabilities, and potential to accelerate creative development workflows. The real benchmark? Turning complex ideas into functional, interactive experiences. #M31Flash #WebGPU #AI

GrowAIHub

79,210 просмотров • 7 дней назад

You can build your 3D models directly from a prompt into Claude code. All thanks to this beautiful python project, build123 a parametric modeling framework for 2D and 3D CAD. It is built on the Open Cascade geometric kernel and it is suitable for 3D printing.
0:36

You can build your 3D models directly from a prompt into Claude code. All thanks to this beautiful python project, build123 a parametric modeling framework for 2D and 3D CAD. It is built on the Open Cascade geometric kernel and it is suitable for 3D printing.

Marco Franzon

20,915 просмотров • 7 месяцев назад

Kimi K3.1 Pro is live on Venice. Kimi.ai’s strongest open-weight model: frontier-level intelligence for long-running coding and agent work, and it even designs 3D models, writing Blender and CAD scripts. Available anonymously. 20% off Pro and Highspeed (API) thru Oct 31.
0:24

Kimi K3.1 Pro is live on Venice. Kimi.ai’s strongest open-weight model: frontier-level intelligence for long-running coding and agent work, and it even designs 3D models, writing Blender and CAD scripts. Available anonymously. 20% off Pro and Highspeed (API) thru Oct 31.

Venice

460,068 просмотров • 7 дней назад

3D AI is leveling up! Rodin 3D AI can create stunning, high-quality 3D models from just text or image inputs. And with its latest update, it can even generate 8K HDRI textures to bring your models to life. Check out the link in the comments!
1:03

3D AI is leveling up! Rodin 3D AI can create stunning, high-quality 3D models from just text or image inputs. And with its latest update, it can even generate 8K HDRI textures to bring your models to life. Check out the link in the comments!

el.cine

46,061 просмотров • 1 год назад

This AI created an interactive 3D textbook !! 💀 I didn't manually create any of these 3d models I built it using: 👉 GPT-6 Astra 👉 Hyper3D Rodin's MCP Hyper3D worked on the 3D Models And GPT-6 Astra made the website with the 3D Models integrated to the UI controls. It even added floating labels for the HEART !! This is what happens when an AI can actually CREATE 3D assets 👇
0:27

This AI created an interactive 3D textbook !! 💀 I didn't manually create any of these 3d models I built it using: 👉 GPT-6 Astra 👉 Hyper3D Rodin's MCP Hyper3D worked on the 3D Models And GPT-6 Astra made the website with the 3D Models integrated to the UI controls. It even added floating labels for the HEART !! This is what happens when an AI can actually CREATE 3D assets 👇

SARAH

135,934 просмотров • 1 месяц назад

Clearly misleading information. I’ve been using Kimi K3.1 Pro since yesterday to procedurally generate 3D models with bpy, and the results have been phenomenal. This is just one example. I’ve generated plenty more, and it does all of this on its own. Not using any 3D Gen model at all. Not sure why everyone is going against Kimi when it's really actually good. It really shouldn’t be that hard to acknowledge that it’s a great model. Besides, it's an open weight model. Doing good for humanity and advancing AI in general
0:33

Clearly misleading information. I’ve been using Kimi K3.1 Pro since yesterday to procedurally generate 3D models with bpy, and the results have been phenomenal. This is just one example. I’ve generated plenty more, and it does all of this on its own. Not using any 3D Gen model at all. Not sure why everyone is going against Kimi when it's really actually good. It really shouldn’t be that hard to acknowledge that it’s a great model. Besides, it's an open weight model. Doing good for humanity and advancing AI in general

The Bugged Dev

20,077 просмотров • 6 дней назад

A Rive customer asked us how to convert their 3D files to vector. So I built this demo to show you don't have to. Instead of baking the 3D into a frame-by-frame vector sequence (the way they'd done it for Lottie in the past), you can keep the real 3D and just make it look vector. The 3D models are standard glTF files, loaded straight into Rive and rendered on the GPU Canvas with flat three-tone shading and outlines. It looks like illustration, but it's still 3D: you can spin it, animate it and relight it, all in real time. The phone screen is a live Rive artboard with real components, layouts, text, data binding and state machines, mapped onto the 3D phone. You can use it in 3D: scroll the app, tap the tabs, it all just works. I built it with the Rive CLI, then pushed it to the Rive Editor, where all of Rive's interactive design and animation tools are there to keep iterating on it. Oh and the whole thing is 344 KB.
0:54

A Rive customer asked us how to convert their 3D files to vector. So I built this demo to show you don't have to. Instead of baking the 3D into a frame-by-frame vector sequence (the way they'd done it for Lottie in the past), you can keep the real 3D and just make it look vector. The 3D models are standard glTF files, loaded straight into Rive and rendered on the GPU Canvas with flat three-tone shading and outlines. It looks like illustration, but it's still 3D: you can spin it, animate it and relight it, all in real time. The phone screen is a live Rive artboard with real components, layouts, text, data binding and state machines, mapped onto the 3D phone. You can use it in 3D: scroll the app, tap the tabs, it all just works. I built it with the Rive CLI, then pushed it to the Rive Editor, where all of Rive's interactive design and animation tools are there to keep iterating on it. Oh and the whole thing is 344 KB.

Guido Rosso

11,410 просмотров • 23 дней назад

THIS STARTUP FORKED BLENDER AND REBUILT IT AROUND AI AGENTS. it's called Mixar. you open a 3D scene and can: → tell an agent what to change directly inside the scene → generate 3D assets from references → retopologize, texture and clean assets → run multiple agents across different scenes → use 400+ AI models from one place they forked Blender itself and are turning it into an agentic 3D platform.
0:30

THIS STARTUP FORKED BLENDER AND REBUILT IT AROUND AI AGENTS. it's called Mixar. you open a 3D scene and can: → tell an agent what to change directly inside the scene → generate 3D assets from references → retopologize, texture and clean assets → run multiple agents across different scenes → use 400+ AI models from one place they forked Blender itself and are turning it into an agentic 3D platform.

matt

85,625 просмотров • 28 дней назад

Hi3D V3.0 is out today, and it is the most detailed and realistic 3D generator I have run. And its first model to build geometry at 2048³ voxels. The materials were the surprise. PBR output is well ahead of V2.1, and fine surface relief survives where the other frontier models smooth it flat. I compared it against the current top models across a wide spread of objects, textures on and textures off. It lands at number 2 in my ranking. For 3D printing I think it takes first outright. Also, they opened 48 hours of free access for the launch, which almost nobody does. Full breakdown:
0:32

Hi3D V3.0 is out today, and it is the most detailed and realistic 3D generator I have run. And its first model to build geometry at 2048³ voxels. The materials were the surprise. PBR output is well ahead of V2.1, and fine surface relief survives where the other frontier models smooth it flat. I compared it against the current top models across a wide spread of objects, textures on and textures off. It lands at number 2 in my ranking. For 3D printing I think it takes first outright. Also, they opened 48 hours of free access for the launch, which almost nobody does. Full breakdown:

Stefan 3D AI

12,920 просмотров • 2 месяцев назад

3D-LLM: Injecting the 3D World into Large Language Models paper page: Large language models (LLMs) and Vision-Language Models (VLMs) have been proven to excel at multiple tasks, such as commonsense reasoning. Powerful as these models can be, they are not grounded in the 3D physical world, which involves richer concepts such as spatial relationships, affordances, physics, layout, and so on. In this work, we propose to inject the 3D world into large language models and introduce a whole new family of 3D-LLMs. Specifically, 3D-LLMs can take 3D point clouds and their features as input and perform a diverse set of 3D-related tasks, including captioning, dense captioning, 3D question answering, task decomposition, 3D grounding, 3D-assisted dialog, navigation, and so on. Using three types of prompting mechanisms that we design, we are able to collect over 300k 3D-language data covering these tasks. To efficiently train 3D-LLMs, we first utilize a 3D feature extractor that obtains 3D features from rendered multi- view images. Then, we use 2D VLMs as our backbones to train our 3D-LLMs. By introducing a 3D localization mechanism, 3D-LLMs can better capture 3D spatial information. Experiments on ScanQA show that our model outperforms state-of-the-art baselines by a large margin (e.g., the BLEU-1 score surpasses state-of-the-art score by 9%). Furthermore, experiments on our held-in datasets for 3D captioning, task composition, and 3D-assisted dialogue show that our model outperforms 2D VLMs. Qualitative examples also show that our model could perform more tasks beyond the scope of existing LLMs and VLMs.
0:50

3D-LLM: Injecting the 3D World into Large Language Models paper page: Large language models (LLMs) and Vision-Language Models (VLMs) have been proven to excel at multiple tasks, such as commonsense reasoning. Powerful as these models can be, they are not grounded in the 3D physical world, which involves richer concepts such as spatial relationships, affordances, physics, layout, and so on. In this work, we propose to inject the 3D world into large language models and introduce a whole new family of 3D-LLMs. Specifically, 3D-LLMs can take 3D point clouds and their features as input and perform a diverse set of 3D-related tasks, including captioning, dense captioning, 3D question answering, task decomposition, 3D grounding, 3D-assisted dialog, navigation, and so on. Using three types of prompting mechanisms that we design, we are able to collect over 300k 3D-language data covering these tasks. To efficiently train 3D-LLMs, we first utilize a 3D feature extractor that obtains 3D features from rendered multi- view images. Then, we use 2D VLMs as our backbones to train our 3D-LLMs. By introducing a 3D localization mechanism, 3D-LLMs can better capture 3D spatial information. Experiments on ScanQA show that our model outperforms state-of-the-art baselines by a large margin (e.g., the BLEU-1 score surpasses state-of-the-art score by 9%). Furthermore, experiments on our held-in datasets for 3D captioning, task composition, and 3D-assisted dialogue show that our model outperforms 2D VLMs. Qualitative examples also show that our model could perform more tasks beyond the scope of existing LLMs and VLMs.

AK

250,112 просмотров • 3 лет назад

Introducing Mint MCP - It turns your coding agent into a 3D engine! ✅ Create Threejs apps and games ✅ Generate 3D models ✅ Rig and animate models ✅ Generate Gaussian Splat worlds ✅ Generate PBR materials ✅ Generate themed Asset packs ✅ Generate audio / sfx ✅ Retopologize 3D models ✅ Convert 3D models to any popular format All the scenes in this demo were created with 5.6 + Mint Mcp + Mint Threejs Skills (links below)
0:38

Introducing Mint MCP - It turns your coding agent into a 3D engine! ✅ Create Threejs apps and games ✅ Generate 3D models ✅ Rig and animate models ✅ Generate Gaussian Splat worlds ✅ Generate PBR materials ✅ Generate themed Asset packs ✅ Generate audio / sfx ✅ Retopologize 3D models ✅ Convert 3D models to any popular format All the scenes in this demo were created with 5.6 + Mint Mcp + Mint Threejs Skills (links below)

tamrat

60,530 просмотров • 3 месяцев назад

Okay, this one's nuts. Hermes Agent can make 3D models now. Meshy's official plugin is live in the Hermes plugin catalog. ▸ Text or a photo to a full 3D model ▸ Texture, rig and animate it ▸ Prep it for 3D printing ▸ Orbit it right in the chat on Hermes Desktop ▸ Export GLB, FBX, OBJ, STL or USDZ If you build games or run a 3D printer, your agent just turned into a 3D studio.
0:27

Okay, this one's nuts. Hermes Agent can make 3D models now. Meshy's official plugin is live in the Hermes plugin catalog. ▸ Text or a photo to a full 3D model ▸ Texture, rig and animate it ▸ Prep it for 3D printing ▸ Orbit it right in the chat on Hermes Desktop ▸ Export GLB, FBX, OBJ, STL or USDZ If you build games or run a 3D printer, your agent just turned into a 3D studio.

Hermes Release Watch

29,868 просмотров • 8 дней назад

Chinese AI models are wiping billions off Big Tech right now. Google just lost $200 billion in a single day, and the model it needed to fight back still isn't ready. Gemini 3.5 Pro, Google's most powerful model, is months behind schedule. Alphabet stock dropped 4.4% that same day. The Deepseek moment is happening again, and the new model is FAR bigger. On the same day Google's delay leaked, a Beijing lab called Moonshot released Kimi K3. It is the largest open model ever built, with 2.8 trillion parameters. It took the number one spot on the Frontend Code Arena, a live coding leaderboard, passing Anthropic's best model. And Moonshot is giving it away for free on July 27. The genius part: Anyone with enough computers can download it and run a frontier level AI without paying a cent to a US company. A single task on Kimi K3 costs about 94 cents. The same work on some American models costs nearly double. So why would a company keep paying premium prices for a model it can now get for free? The entire US AI business is built on selling access to models that cost billions to train. If a free Chinese version does most of the same work, that pricing power starts to crack. And Kimi is close to the best. On one closely watched intelligence ranking it scored 57, just behind the top American models GPT-5.6 Sol and Fable 5, and ahead of Claude Opus 4.8. Bank of America told clients that Kimi proves Chinese labs can keep making big leaps even with limited chips. And the founder of Moonshot, Yang Zhilin, learned to build AI as a researcher INSIDE Google. Google literally wrote the 2017 paper that made all of these models possible. Now the people who studied its work are using it to destroy Google, and handing it out for free. What happens next: Kimi K3's weights go public on July 27. Google reports earnings on July 22, and everyone will be asking the same question about Gemini. If free models keep topping the charts, every valuation built on paid AI access has to be rewritten. What do you think?
1:10

Chinese AI models are wiping billions off Big Tech right now. Google just lost $200 billion in a single day, and the model it needed to fight back still isn't ready. Gemini 3.5 Pro, Google's most powerful model, is months behind schedule. Alphabet stock dropped 4.4% that same day. The Deepseek moment is happening again, and the new model is FAR bigger. On the same day Google's delay leaked, a Beijing lab called Moonshot released Kimi K3. It is the largest open model ever built, with 2.8 trillion parameters. It took the number one spot on the Frontend Code Arena, a live coding leaderboard, passing Anthropic's best model. And Moonshot is giving it away for free on July 27. The genius part: Anyone with enough computers can download it and run a frontier level AI without paying a cent to a US company. A single task on Kimi K3 costs about 94 cents. The same work on some American models costs nearly double. So why would a company keep paying premium prices for a model it can now get for free? The entire US AI business is built on selling access to models that cost billions to train. If a free Chinese version does most of the same work, that pricing power starts to crack. And Kimi is close to the best. On one closely watched intelligence ranking it scored 57, just behind the top American models GPT-5.6 Sol and Fable 5, and ahead of Claude Opus 4.8. Bank of America told clients that Kimi proves Chinese labs can keep making big leaps even with limited chips. And the founder of Moonshot, Yang Zhilin, learned to build AI as a researcher INSIDE Google. Google literally wrote the 2017 paper that made all of these models possible. Now the people who studied its work are using it to destroy Google, and handing it out for free. What happens next: Kimi K3's weights go public on July 27. Google reports earnings on July 22, and everyone will be asking the same question about Gemini. If free models keep topping the charts, every valuation built on paid AI access has to be rewritten. What do you think?

Ricardo

48,123 просмотров • 3 месяцев назад

This first-person shooter was built from a single prompt, using 404-generated 3D assets. On the latest Novelty Search, 404 explains how SN17 is moving from Titan, its open image-to-3D model built with miners, toward 3D assets generated as code that AI agents can edit and use directly in games. That work is part of a broader direction for 404: building world models where AI agents can learn and act. Hosted by const Full episode in the first comment
1:08

This first-person shooter was built from a single prompt, using 404-generated 3D assets. On the latest Novelty Search, 404 explains how SN17 is moving from Titan, its open image-to-3D model built with miners, toward 3D assets generated as code that AI agents can edit and use directly in games. That work is part of a broader direction for 404: building world models where AI agents can learn and act. Hosted by const Full episode in the first comment

Openτensor Foundaτion

19,186 просмотров • 2 месяцев назад

Gemini 3.5 Flash is a step forward for Google on speed and agentic capabilities but comes at a trade-off of being higher cost than prior models We have measured up to ~280 output tokens/sec, placing it on the speed/intelligence Pareto frontier and well ahead of Gemini 3 Flash. It also shows a major uplift on agentic tasks, reaching ~1650 ELO on GDPVal-AA. The trade-off: cost is up ~5x versus Gemini 3 Flash, driven by higher token prices (3x higher than Gemini 3 Flash) and higher token usage. In this video, Declan Jackson, Member of Technical Staff at Artificial Analysis, breaks it down.
1:57

Gemini 3.5 Flash is a step forward for Google on speed and agentic capabilities but comes at a trade-off of being higher cost than prior models We have measured up to ~280 output tokens/sec, placing it on the speed/intelligence Pareto frontier and well ahead of Gemini 3 Flash. It also shows a major uplift on agentic tasks, reaching ~1650 ELO on GDPVal-AA. The trade-off: cost is up ~5x versus Gemini 3 Flash, driven by higher token prices (3x higher than Gemini 3 Flash) and higher token usage. In this video, Declan Jackson, Member of Technical Staff at Artificial Analysis, breaks it down.

Artificial Analysis

16,055 просмотров • 5 месяцев назад

Kimi. Kimi. Kimi. 🔥 You can literally build a beautiful website in a couple of hours using Kimi.ai 3D effects, images, UI elements, everything gets generated on its own. You don't have to spend time finding assets, fonts, or any other design resources. This is the kind of intelligence I want from AI. Give it a prompt and a reference, and it should figure out the rest. Generate the assets, coordinate the sub agents, and ship a complete experience. That's the future. Live: Code: Design:
0:32

Kimi. Kimi. Kimi. 🔥 You can literally build a beautiful website in a couple of hours using Kimi.ai 3D effects, images, UI elements, everything gets generated on its own. You don't have to spend time finding assets, fonts, or any other design resources. This is the kind of intelligence I want from AI. Give it a prompt and a reference, and it should figure out the rest. Generate the assets, coordinate the sub agents, and ship a complete experience. That's the future. Live: Code: Design:

The Bugged Dev

31,207 просмотров • 2 месяцев назад