AI Models
Image, video and 3D generation models and the tools built on them.
AI imageMistral and NVIDIA Link Physics AI to Engineering Simulations
Mistral and NVIDIA plan to combine physics AI and 3D simulation tools in a workflow spanning design, simulation, and AI training. Specific use cases for architecture and archviz have not yet been announced.
OpenAI Models Are Coming to Google Cloud Model Garden, Alongside Codex
OpenAI and Google Cloud will bring OpenAI models and Codex workflows to Google Cloud Model Garden. Enterprise customers will be able to make purchases through Marketplace, with general availability expected to begin in the coming weeks.
NVIDIA Agent Toolkit expands with a new enterprise data layer
NVIDIA has added Agent Knowledge Foundation to Agent Toolkit, aiming to turn company data into information that agents can work with. The update brings together data, models and security components.
NVIDIA Expands Its Open Model Portfolio for Europe: Nemotron 3.5 Unveiled
On October 21, NVIDIA announced open models for a range of fields, including Nemotron 3.5 Super VL, Kumo and Cosmos. Organizations can customize the models with their own data and run them on infrastructure they control.
AI imageAnthropic and Nscale in Talks for a New $45 Billion Deal
Anthropic and AI infrastructure provider Nscale are reportedly discussing a new deal of the same size, following the $45 billion agreement announced more than a month ago. Talks have not yet concluded.
AI imageMiniMax M3.1 Flash launches with lower task costs and strong reasoning
MiniMax announced the M3.1 Flash model on October 19. Now available through the company’s API, it outperformed the previous M3 in tests while bringing the cost per task down to around $0.22.
AI imageKling 4.0 is here: 30-second video, stereo audio and a Flash version
Kling AI announced version 4.0 on 19 October 2026. The update offers videos up to 30 seconds per generation, more image and video references, stereo audio and multilingual lip-sync.
AI imageNVIDIA unveils Nemotron 3.5 models for enterprise agents and local use
On 15 October, NVIDIA announced the 120-billion-parameter Nemotron 3.5 Super VL for enterprise agents, along with two Bolt models for local use. Two models are now available, while Bolt E4B is expected later this month.
AI imageDeepSeek V4.1 Flash-1017 launches, targeting coding and agent tasks
DeepSeek made the V4.1 Flash-1017 update available on 17 October through its website, app and API. The company says it offers faster performance on coding and agent tasks, shorter outputs and new search APIs.
AI imageAMD to Acquire World Labs in $8.2 Billion All-Stock Deal
According to Startup Fortune, AMD announced on September 28 a plan to acquire San Francisco-based World Labs in an all-stock deal valued at approximately $8.2 billion. The deal has not yet closed.
AI imageGlimpse launches: A new model designed to reason over images
Elorian released Glimpse, its first model focused on visual reasoning, as a research preview on October 16. The model is available via API, while its chat interface is waitlist-only.
AI imageGemini 4 Argon scores 97.5% in provider-specific ARC-AGI-3 test
Google DeepMind’s Gemini 4 Argon scored 97.5% on ARC-AGI-3 in a provider-specific test setup. Its score fell to 49.8% at the low reasoning level on the Standard Harness.
AI imageBeam Released: 501 Billion Parameters and a 1 Million-Token Context Window
Reflection’s 501-billion-parameter Beam model was released on Hugging Face on 16 October 2026 under the Apache 2.0 license. The text-focused model supports a 1 million-token context window; running it on your own infrastructure requires high-memory hardware.
AI imageMuse Spark Open unveiled: A multimodal model with 816 billion parameters
Meta’s Muse Spark Open can process image, audio and video inputs alongside text. With approximately 816 billion total parameters, the model focuses on coding tasks and requires high-memory hardware.
FLUX.3 Image Klein Launches: Layout-Controlled Generation Comes to Diffusers
Black Forest Labs’ FLUX.3 Image Klein model became available on Hugging Face Diffusers on 15 October. Alongside generation from text and multiple reference images, it lets users define scene element positions with bounding boxes.
NVIDIA Nemotron 3.5 Super VL launches: an open model with 120 billion parameters
NVIDIA announced the 120-billion-parameter Nemotron 3.5 Super VL and two Bolt models. Super VL and Bolt 74B-A9B are now available, while Bolt E4B is scheduled to launch on 23 October.
AI imageLFM2.5-24B-A2B launches: A local agent model with 2B active parameters
Liquid AI announced the LFM2.5-24B-A2B model on 13 October. With 24 billion total parameters and 2 billion active parameters per token, it targets locally run, tool-using agents.
Gemini 4 Argon unveiled: 3D simulation workflows for robotics
Google DeepMind introduced Gemini 4 Argon on 13 October 2026 for robotics and physical reasoning. The announcement includes examples of building simulation scenes in MuJoCo from real-world footage.
Gemini 4 Argon launches: Google demonstrates 3D scene creation with Blender
Google has made Gemini 4 Argon available globally. The company shared examples of the model creating 3D scenes with Antigravity and Blender, and announced introductory API pricing.
AI imageMeta Opens Access to Muse Wonder and Wonder Full Models
Meta has made its Muse Wonder and Muse Wonder Full models available to Muse Code subscribers; Wonder is also offered through the Meta Model API. The company plans to use Wonder in its Muse agent but has not provided an integration timeline.
AI imageBlueprints AI aims to launch its second-generation version this week
Blueprints AI plans to make its second-generation software available this week. The tool can create architectural and engineering drawings from a sketch or project brief. The company says changes will be reflected throughout the project and checked against local building codes.
AI imageOdyssey-3 opens to the public: Real-time interactive worlds from text
Odyssey has made its Odyssey-3 world model available as a public research preview. The free online demo generates real-time environments from text prompts; developers must apply for API access.
AI imageAi2’s GPU Scheduler Budgets Time for Research Projects
On October 9, Ai2 described a scheduler that manages GPU time for research clusters using project budgets and fair-sharing rules. The organization reports a 74% reduction in repairs requiring human intervention.
AI imageNVIDIA Omniverse: Simulation Examples Featuring Astra Agents Shared
NVIDIA shared AI-powered simulation examples developed with Omniverse libraries. Projects ranging from warehouse robot testing to digital twins show how agents are used under developer guidance.
About this category
In the AI Models category, we cover developments in AI models that generate images, video, and 3D content, with a focus on architecture and visualization. Here you’ll find new releases, pricing and availability updates, and features that are genuinely useful for architectural rendering from models such as GPT Image, Nano Banana, Seedream, FLUX, Veo, and Kling.
We compile each story from official announcements and reliable industry sources, and consider what it means for architects and visualization artists. This category also features our monthly Model League results, comparing models with our own test set, as well as workflows for rapid AI-assisted concept development, material variations, and turning images into video.
AI models change quickly, so we clearly state each model’s version, release date, and access terms (API, subscription, or open weights). We also note practical details such as commercial usage rights, availability in Türkiye, and Turkish-language prompt support.