AI Models
•
July 27, 2026
•
15 min read
Every AI Company, Model Series, and Latest Flagship Release (2026 Directory)
A comprehensive, SEO-indexed directory of every major AI company worldwide, their active model lineages, and flagship releases across text, vision, audio, video, and physical AI.
Mohid Mirza
Co-Founder of AcceleratedLogic AI
## Introduction: The 2026 AI Frontier Landscape
In this definitive 2026 directory, we index every major company developing artificial intelligence models, including their active model lineages, architectural paradigms, and latest flagship releases across LLMs, vision, speech, video, and physical AI.
Note: In accordance with directory standards, model series without active releases for over six months (such as GPT-OSS) have been omitted. If you know of an AI lab or model release missing from this list, please contact us at [email protected].
## Global Frontier Tech Giants
### OpenAI
**Model Series:** ChatGPT, GPT Image, GPT-Live
**Latest Models:** ChatGPT 5.6 Luna/Terra/Sol, GPT Image 2, GPT-Live-1, GPT-Live-1 mini
OpenAI remains a dominant force in frontier artificial intelligence. Beyond their flagship ChatGPT text and reasoning series, OpenAI develops real-time multimodal audio models (GPT-Live) and next-generation image synthesis systems (GPT Image 2).
### Anthropic
**Model Series:** Claude Haiku, Claude Sonnet, Claude Opus, Claude Fable
**Latest Models:** Haiku 4.5, Sonnet 5, Opus 5, Fable 5
Anthropic is dedicated to Constitutional AI and alignment research. Currently, Anthropic produces some of the most capable and reasoning-dense models on Earth, spanning the lightweight Haiku series, workhorse Sonnet, benchmark-topping Opus 5, and flagship Fable 5.
### Google
**Model Series:** Gemini, Gemma, VEO (Video), Omni (Video), Gemini Image, Lyria (Music), Gemini TTS
**Latest Models:** Gemini 3.5 Flash-lite, Gemini 3.6 Flash, Gemini 3.1 Pro, Gemma 4 E2B, Gemma 4 E4B, Gemma 4 12B, Gemma 4 26B A4B, Gemma 4 31B, VEO 3.1, Omni, Omni Flash, Gemini 3.1 Flash Image, Gemini 3.1 Flash-lite Image, Gemini Pro Image, TTS 3.1
Google develops models across every major AI modality. Their Gemini series excels at long-context comprehension, ultra-low-cost question answering, and multimodal reasoning. Their open-weights Gemma 4 series offers state-of-the-art coding and reasoning efficiency for on-device and cloud deployments.
### SpaceXAI
**Model Series:** Grok, Grok Imagine, Grok Imagine Video
**Latest Models:** Grok 4.5, Grok Imagine 1.5, Grok Imagine Video 1.5
SpaceXAI develops the Grok lineup. Recognized for rapid reasoning, strong agentic tool execution, and fewer safety restrictions on real-world discourse.
### Nvidia
**Model Series:** Nemotron
**Latest Model:** Nemotron 3 Ultra
While primarily the world's leading AI chipmaker, Nvidia publishes open-source foundation models under the Nemotron series, optimized for enterprise pipelines, synthetic data generation, and GPU acceleration.
### Meta
**Model Series:** Muse Spark, Muse Image, Muse Video
**Latest Models:** Muse Spark 1.1, Muse Image, Muse Video
Meta continues to advance open intelligence with its new Muse family, spanning text reasoning (Muse Spark 1.1), image generation, and high-fidelity video synthesis.
### Microsoft AI (MAI)
**Model Series:** MAI-Thinking, MAI-Code, MAI-Image, MAI-Voice, MAI-Transcribe
**Latest Models:** MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5, MAI-Voice-2, MAI-Voice-2-Flash, MAI-Transcribe-1.5
MAI is Microsoft's in-house AI research division. Their foundation models are deeply embedded across Windows and Azure ecosystems, with industry-leading strengths in visual generation and voice transcription.
### Amazon
**Model Series:** Nova
**Latest Models:** Nova 2 Lite, Nova 2 Pro (Preview), Nova 2 Sonic, Nova Multimodal Embeddings
Amazon's Nova family focuses on fast, cost-effective multimodal models for text, image, video, real-time speech (Sonic), and embeddings deployed via Amazon Bedrock.
### Apple
**Model Series:** Apple Foundation Models (AFM)
**Latest Models:** AFM 3 family (Core, Core Advanced, Cloud variants)
On-device and Private Cloud Compute foundation models powering Apple Intelligence, engineered specifically for Apple Silicon, user privacy, and OS integration.
## The Chinese AI Ecosystem & Cost-Efficiency Leaders
### DeepSeek
**Model Series:** DeepSeek
**Latest Models:** DeepSeek V4 Flash / Pro
DeepSeek is internationally recognized for pioneering hyper-efficient Mixture-of-Experts architectures, setting global standards for inference cost-efficiency and mathematical reasoning.
### Alibaba
**Model Series:** Qwen, Qwen Image, Wan
**Latest Models:** Qwen 3.7-Plus, Qwen 3.8-Max, Qwen 3.6 35B A3B, Qwen 3.6 27B, Qwen Image 2.0, Wan 2.7
Alibaba's Qwen series bridges open-source and proprietary models, providing extreme token efficiency, superior coding performance, and competitive image/video models (Wan 2.7).
### Zhipu AI
**Model Series:** GLM, GLM-Image
**Latest Models:** GLM 5.2, GLM-Image
Zhipu AI produces the GLM family—cost-effective yet exceptionally powerful foundation models competing at the highest tiers of global benchmarks.
### Moonshot AI
**Model Series:** Kimi
**Latest Models:** Kimi K3
Moonshot AI builds the Kimi lineup. Kimi K3 leads Chinese AI models in autonomous software engineering, long-context reasoning, and complex coding.
### MiniMax
**Model Series:** MiniMax
**Latest Model:** MiniMax M3
MiniMax is a major Chinese AI laboratory producing highly capable, large-scale multimodal foundation models for enterprise and consumer deployment.
### Kuaishou (KwaiKAT)
**Model Series:** KAT-Coder, KAT-Dev, Kling
**Latest Models:** KAT-Coder-Pro V2.5, KAT-Coder-Air V2.5, Kling AI 3.0
Kuaishou's KwaiKAT team released KAT-Coder-Pro V2.5—China's first agentic coding model capable of end-to-end software engineering. They also develop Kling AI, a world-class video generation platform.
### Tencent
**Model Series:** Hunyuan (Hy), Hunyuan Video
**Latest Models:** Hy3, Hunyuan Video
Tencent's Hunyuan team builds Apache 2.0 open Mixture-of-Experts models for long-context agentic tasks alongside Hunyuan Video for scalable production video generation.
### InclusionAI (Ant Group)
**Model Series:** Ling, Ring, Ming
**Latest Models:** Ling-2.6-1T, Ling-2.6-flash, Ling-3.0-flash
Ant Group's AI division develops the Ling family (high-efficiency text), Ring (explicit reasoning), and Ming (multimodal), featuring the 1-trillion parameter Ling-2.6-1T flagship.
### Baidu
**Model Series:** ERNIE
**Latest Model:** ERNIE 5.1
Baidu's ERNIE 5.1 compresses parameter footprint to 1/3 of ERNIE 5.0 while dramatically expanding agentic tool execution and reasoning capabilities.
### ByteDance
**Model Series:** Doubao/Seed, Seedance
**Latest Models:** Seed 2.1 Pro/Turbo, Seedance
ByteDance's Seed models drive autonomous agentic execution, while their Seedance video model generates roughly $2 billion in annual recurring revenue.
### StepFun
**Model Series:** Step
**Latest Model:** Step 3.7 Flash
StepFun produces Apache 2.0 open sparse Mixture-of-Experts vision-language models designed specifically for coding agents, tool invocation, and search workflows.
### Xiaomi
**Model Series:** MiMo
**Latest Models:** MiMo-V2.5, MiMoV2.5-Pro
Xiaomi produces the MiMo foundation series, engineered for high throughput, low latency, and cost-efficient edge/cloud inference.
### LongCat (Meituan)
**Model Series:** LongCat
**Latest Model:** LongCat-2.0
Meituan's in-house 1.6-trillion parameter MoE model (48B active per token) features a native 1-million token context window, trained on domestic hardware for agentic software engineering.
### 01.AI
**Model Series:** Yi
**Latest Models:** Yi-Lightning, Yi series
Founded by Kai-Fu Lee, 01.AI produces high-performance bilingual open MoE models optimized for inference speed.
### Shanghai AI Laboratory
**Model Series:** InternLM, InternVL, Intern-S
**Latest Models:** InternLM3, Intern-S1
Major Chinese research institution releasing open multimodal, mathematical, and scientific reasoning foundation models.
### Huawei
**Model Series:** Pangu
**Latest Models:** Pangu 5.5, openPangu
Huawei's Pangu foundation series focuses on enterprise, industrial, weather forecasting, and scientific research applications.
## Emerging Frontier Labs & Open Intelligence
### Agnes AI
**Model Series:** Agnes Flash, Agnes Pro, Agnes Image, Agnes Video
**Latest Models:** Agnes 2.5 Flash, Agnes 2.5 Pro Alpha, Agnes Image 2.1 Flash, Agnes Video V2.0
Agnes AI is a Singapore-based AI lab that trains its own full-modality foundation models in-house across text, image, and video, and offers them through a free omni-modal API that has passed 3 million users. The company built its reputation on removing cost as a barrier — its Agnes 2.0 model series processed 5.41 trillion tokens combined in a single week. Currently, Agnes AI's lineup spans the free, coding-and-agent-focused Flash series, the paid, reasoning-dense Pro Alpha line for deeper multi-step tasks, and dedicated Image and Video generation models.
### Thinking Machines Lab
**Model Series:** Inkling
**Latest Model:** Inkling 1.0 (975B MoE Open Weights)
Founded by former OpenAI CTO Mira Murati, Thinking Machines Lab publishes open-weight foundation models and enterprise fine-tuning platforms.
### Poolside
**Model Series:** Laguna
**Latest Model:** Laguna S 2.1 (118B Open MoE)
Poolside develops open-weight agentic coding models trained in execution environments with reinforcement learning.
### PrismML
**Model Series:** Bonsai
**Latest Model:** Bonsai 27B (1-bit / Ternary On-Device)
PrismML specializes in extreme low-bit quantization, enabling 27B parameter models to execute locally on consumer smartphones and laptops.
### Cohere
**Model Series:** Command, North, Rerank
**Latest Models:** Command A6, North 1.5, Rerank 3.5
Cohere delivers enterprise RAG, search, and agentic reasoning models optimized for enterprise data pipelines and multi-step tool execution.
### ElevenLabs
**Model Series:** Eleven, Scribe
**Latest Models:** Eleven Multilingual v3, Scribe 2
ElevenLabs leads audio AI in voice synthesis, real-time speech conversion, dubbed media processing, and automated transcription.
### Black Forest Labs
**Model Series:** FLUX
**Latest Models:** FLUX.1.1 Pro, FLUX.1 Kontext
Black Forest Labs produces state-of-the-art open and commercial image generation and editing foundation models.
## Comprehensive Directory Summary Table
| Company | Origin | Model Lineage | Flagship Capability / Focus |
|---|
|---|---|---|---|
| OpenAI | USA | ChatGPT, GPT Image, GPT-Live | Frontier Reasoning, Multimodal Real-time Audio |
|---|
| Anthropic | USA | Claude (Haiku, Sonnet, Opus, Fable) | Constitutional AI, Elite Agentic Coding & Knowledge Work |
|---|
| USA | Gemini, Gemma, VEO, Omni | 1M+ Context, High Speed, Multimodal Vision/Speech/Video |
|---|
| SpaceXAI | USA | Grok, Grok Imagine | Fast Tool Execution, Uncensored Real-time Web Context |
|---|
| Nvidia | USA | Nemotron | Enterprise Synthetic Data & Acceleration |
|---|
| Meta | USA | Muse Spark, Muse Image, Muse Video | Open-Leaning Multimodal Intelligence |
|---|
| Microsoft AI | USA | MAI-Thinking, MAI-Code, MAI-Voice | OS-Integrated Reasoning & Speech Synthesis |
|---|
| Amazon | USA | Nova | Bedrock-Integrated Multimodal & Real-time Voice |
|---|
| Apple | USA | AFM (Apple Foundation Models) | On-Device & Private Cloud Compute Intelligence |
|---|
| DeepSeek | China | DeepSeek V4 | Hyper-Efficient Open MoE Architecture |
|---|
| Alibaba | China | Qwen, Wan | High-Efficiency Open LLMs & Video Generation |
|---|
| Zhipu AI | China | GLM | Enterprise Open-Weight Cost Leaders |
|---|
| Moonshot AI | China | Kimi K3 | 2.8T Parameter Open MoE, 1M Context |
|---|
| MiniMax | China | MiniMax M3 | Multimodal Enterprise Foundation Models |
|---|
| Kuaishou | China | KAT-Coder, Kling AI | Agentic Software Engineering & Video Generation |
|---|
| Tencent | China | Hunyuan (Hy) | Apache 2.0 Open MoE & Video Models |
|---|
| InclusionAI | China | Ling, Ring, Ming | 1-Trillion Parameter Enterprise MoE |
|---|
| Baidu | China | ERNIE | Efficient Enterprise Agentic Workflows |
|---|
| ByteDance | China | Seed, Seedance | High-Scale Commercial Video & Autonomous Agents |
|---|
| StepFun | China | Step | Open Sparse MoE VLM for Search & Agents |
|---|
| Xiaomi | China | MiMo | Ultra-Low-Latency Edge/Cloud Inference |
|---|
| Meituan | China | LongCat | Domestic Hardware Trained Agentic MoE |
|---|
| Agnes AI | Singapore | Agnes (Flash, Pro, Image, Video) | Full-Modality Free Omni-Modal API & Budget Reasoning |
|---|
| Thinking Machines | USA | Inkling | Open 975B MoE Base Models for Fine-tuning |
|---|
| Poolside | USA | Laguna | RL Code Execution, Open 118B Agent Models |
|---|
| PrismML | USA | Bonsai | 1-bit & Ternary On-Device Model Compression |
|---|
| Cohere | USA | Command, North, Rerank | Enterprise RAG, Agentic Workflows, Citation Grounding |
|---|
| ElevenLabs | USA | Eleven, Scribe | Ultra-realistic Voice, Audio, Dubbing |
|---|
| Black Forest Labs | Germany | FLUX | State-of-the-art Image & Video Synthesis |
|---|