← Back to Articles Directory
Always Updating Lists • August 1, 2026 • 12 min read

Every AI Company, Model Series, and Latest Flagship Release (2026 Guide)

A comprehensive research overview 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 & Lead Programmer of AcceleratedLogic AI

Introduction: The 2026 AI Frontier Landscape

In this definitive 2026 research overview, we analyze 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. This guide is continuously updated as new frontier, open-weight, and specialized agentic models drop.
Note: 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 Sol / Terra / Luna, 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 (GPT-5.6 Sol, Terra, and Luna), OpenAI develops real-time multimodal audio models (GPT-Live) and next-generation image synthesis systems (GPT Image 2).

Anthropic

Model Series: Claude (Haiku, Sonnet, Opus, Fable) Latest Models: Claude 5 Opus, Claude Fable 5, Claude Sonnet 5, Claude Haiku 4.5
Anthropic is dedicated to Constitutional AI and alignment research. Anthropic produces some of the most capable models on Earth; their Claude 5 Opus release sits at #1 on the Artificial Analysis Intelligence Index (scoring 61) and leads global agentic software benchmarks while cutting operational costs by 50% compared to Fable 5.

Google

Model Series: Gemini, Gemma, VEO (Video), Omni (Video), Gemini Image, Lyria (Music), Gemini TTS Latest Models: Gemini 3.6 Flash, Gemini 3.5 Flash-lite, Gemini 3.1 Pro, Gemma 4 (31B, 26B, 12B, E4B, E2B), VEO 3.1, Omni Flash, TTS 3.1
Google develops models across every major AI modality: - Gemini 3.6 Flash: High-efficiency flagship model for coding, agentic workflows, web/app development (1.05M context window; $1.50/M input, $7.50/M output standard; $0.75/M input, $3.75/M output batch). Ranks prominently across Academia (#31), Finance (#39), Health (#33), Legal (#18), and Marketing (#38). - Gemini 3.5 Flash Lite: Ultra-fast subagent execution model for multi-agent systems (1.05M context window; $0.30/M input, $2.50/M output standard; $0.15/M input, $1.25/M output batch). - Gemma 4: Open-weights family (31B, 26B, 12B, E4B, E2B) for on-device and cloud deployment.

SpaceXAI / xAI

Model Series: Grok, Grok Imagine, Grok Imagine Video Latest Models: Grok 4.6, Grok 4.5, Grok Imagine 1.5, Grok Imagine Video 1.5
SpaceXAI develops the Grok lineup. Their newest flagship, Grok 4.6, features a 500K token context window, $2/M input & $6/M output pricing, and scores 61 on the Artificial Analysis Intelligence Index with state-of-the-art vector SVG generation and interactive 3D WebGL coding performance.

Nvidia

Model Series: Nemotron Latest Models: Nemotron 3.5 Lightning, Nemotron 3 Ultra
Nvidia publishes open-source foundation models under the Nemotron series. Nemotron 3.5 Lightning is a 30B parameter MoE model (3B active) engineered for high-throughput, low-latency agent orchestration and synthetic data generation.

Meta

Model Series: Muse Spark, Muse Glimmer, Muse Image, Muse Video Latest Models: Muse Spark 1.2, Muse Spark 1.1, Muse Glimmer 30B, Muse Image, Muse Video
Meta continues to advance open intelligence with its Muse family: - Muse Spark 1.2 / 1.1: Multimodal reasoning models accepting text, images, video, audio, and PDFs with a 1.05M context window ($1.25/M input, $4.25/M output; 54 Artificial Analysis Intelligence score). - Muse Glimmer 30B: 29.6B dense open-source model (Apache 2.0) with a 131K context window, achieving remarkable complex discrete math reasoning under Q2 quantization.

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 V4 Latest Models: DeepSeek V4 Pro 0813, DeepSeek V4 Flash 0731
DeepSeek leads global cost-efficiency in Mixture-of-Experts architectures: - DeepSeek V4 Pro 0813: 1.6T parameter MoE (49B active per token), 1M token context window, priced at $0.435/M input and $0.87/M output. - DeepSeek V4 Flash 0731: 284B parameter MoE (13B active per token), 1M token context window, open weights (MIT), priced at $0.07–$0.14/M input and $0.14–$0.28/M output ($0.03 average cost per task).

Alibaba

Model Series: Qwen, Qwen Image, Wan Latest Models: Qwen 3.8-Max, Qwen 3.7-Plus, 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: - Qwen 3.8-Max: 2.4T MoE (95B active per token), 1M token context window, 92.2% GPQA Diamond score, flat $2.00/M input and $6.00/M output pricing. - Qwen 3.7-Plus: High-speed agentic reasoning model with 1M context window ($0.40/M input, $1.20/M output).

InclusionAI (Ant Group)

Model Series: Ling, Ring, Ming Latest Models: Ling-3.0-flash, Ling 3.0 Tiny
InclusionAI (Ant Group division) offers two extreme archetypes: - Ling-3.0-flash: 124B parameter MoE (5.1B active per token), 262K context window, #1 Intelligence Index score (37) at 333.1 tokens/sec throughput ($0.021/M input & $0.063/M output promo rate; $0.075 / $0.22 standard). Targeted at production-scale agentic inference, SEO (#47), and Translation (#47). - Ling 3.0 Tiny: 7.9B parameter MoE (1.3B active per token), 262K context window, $0.00/M free hosted API, locally runnable on consumer GPUs (12–16GB VRAM).

Moonshot AI

Model Series: Kimi Latest Models: Kimi K3
Moonshot AI's Kimi K3 is a 2.8-trillion parameter open-weight multimodal reasoning model (104B active per token) with a 1.05M context window ($2.80/M input, $14.00/M output). Architecture utilizes hybrid KDA linear attention + Gated MLA, AttnRes depth connections, and Quantile Balancing for load distribution. Ranks high in Academia (#19), Finance (#12), Legal (#32), Marketing (#28), and SEO (#42).

LongCat (Meituan)

Model Series: LongCat Latest Model: LongCat 2.0
Meituan's LongCat 2.0 is a sparse Mixture-of-Experts language model featuring 1.6T total parameters (48B active per token) and a 1.05M context window ($0.30/M input, $1.20/M output after 60% promo). Built for repository-level coding, long-horizon problem solving, and agentic workflows.

Kuaishou (Kwaipilot / 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 Kwaipilot team develops flagship agentic coding models capable of receiving entire business workflows and autonomously locating and executing modifications across full codebases: - KAT-Coder-Pro V2.5: 256K context window ($0.74/M input, $2.96/M output). - KAT-Coder-Air V2.5: 256K context window ($0.15/M input, $0.60/M output). - Kling AI 3.0: World-class video generation platform.

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. GLM 5.2 famously proved vital in active cybersecurity incident response at Hugging Face.

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.

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 (Hy3) for long-context agentic tasks alongside Hunyuan Video for scalable production video generation.

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.

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, Specialized Speed Demons, Routers & Open Intelligence

Thinking Machines Lab

Model Series: Inkling Latest Models: Inkling 1.0, Inkling Small
Founded by former OpenAI CTO Mira Murati, Thinking Machines Lab publishes open-weight multimodal MoE models: - Inkling 1.0: 975B MoE (41B active per token), 1.05M context window ($0.95/M input, $4.05/M output standard; $0.50/M input, $2.025/M output batch). Features native text, image, and audio understanding. - Inkling Small: 276B MoE (12B active per token), 524K/1M context window, Apache 2.0 open weights ($0.30/M input, $1.20/M output).

Poolside

Model Series: Laguna Latest Models: Laguna S 2.1, Laguna S 2.1 (Free)
Poolside develops open-weight agentic coding models trained in execution environments with reinforcement learning: - Laguna S 2.1 (Free): 118B total MoE (8B active per token), 262K context window, $0/M input and output tokens under OpenMDW-1.1 license (70.2% Terminal-Bench 2.1, 40.4% DeepSWE). - Laguna S 2.1: 118B total MoE (8B active per token), 1.05M context window ($0.09/M input, $0.18/M output after 10% discount).

OpenRouter

Model Series: Auto Router, Pareto Code Latest Feature: Auto Router (Beta)
OpenRouter's Auto Router (Beta) is a task-aware dynamic request router with a 2M token context window. It classifies incoming prompts and dynamically routes them to the most effective model based on user-configured cost-quality tradeoffs.

Celeris Labs

Model Series: Celeris Latest Model: Celeris-1
Celeris Labs is an artificial intelligence research lab focused on building ultra-fast LLMs. Their flagship Celeris-1 model abandons traditional autoregressive token generation for a novel diffusion architecture, delivering sub-200ms real-time latency and ~1,500+ tokens/sec output throughput while achieving 75.9% on MMLU-Pro.

Agnes AI (Sapiens AI)

Model Series: Agnes Flash, Agnes Pro, Agnes Image, Agnes Video Latest Models: Agnes 2.5 Pro Alpha, Agnes 2.5 Flash, 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 across text, image, and video. Their Agnes 2.5 Pro Alpha model pairs mid-tier reasoning (32% HLE, 88% GPQA Diamond) with aggressively low $0.45 / $0.90 per 1M token pricing ($0.18 blended).

Motif Technologies

Model Series: Motif Latest Model: Motif 3
South Korean AI startup Motif Technologies has released Motif 3, a 314-billion parameter homegrown MoE open-source model designed to compete directly with Chinese open-weights models.

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 complex local reasoning 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 Release Summary Table

Company Origin Model Lineage Active Model & Specs Context Input Price (1M) Output Price (1M) Core Focus / Category
OpenAI USA ChatGPT, GPT-Live ChatGPT 5.6 Sol / Terra / Luna 1M $2.50 - $15.00 $7.50 - $45.00 Frontier Reasoning & Multimodal
Anthropic USA Claude Claude 5 Opus / Fable 5 / Sonnet 5 1M $3.00 - $15.00 $15.00 - $75.00 #1 Intelligence Index (61), Agentic Code
Google USA Gemini, Gemma Gemini 3.6 Flash 1.05M $1.50 ($0.75 batch) $7.50 ($3.75 batch) High-Speed Coding, Agents, Multimodal
Google USA Gemini Gemini 3.5 Flash Lite 1.05M $0.30 ($0.15 batch) $2.50 ($1.25 batch) Subagent Task Execution
SpaceXAI / xAI USA Grok Grok 4.6 500K $2.00 $6.00 61 AA Score, Vector SVG & 3D WebGL
DeepSeek China DeepSeek V4 DeepSeek V4 Pro 0813 (1.6T MoE, 49B Active) 1M $0.435 $0.87 Elite Reasoning, $0.435/$0.87 Ultra-Cheap
DeepSeek China DeepSeek V4 DeepSeek V4 Flash 0731 (284B MoE, 13B Active) 1M $0.07 - $0.14 $0.14 - $0.28 Workhorse MoE, Open MIT Weights
InclusionAI China Ling Ling-3.0-flash (124B MoE, 5.1B Active) 262K $0.021 ($0.075 std) $0.063 ($0.22 std) #1 Intelligence Index (37), 333 t/s
InclusionAI China Ling Ling 3.0 Tiny (7.9B MoE, 1.3B Active) 262K $0.00 (Free API) $0.00 (Free API) Local GPU (12-16GB VRAM), Free API
Poolside USA Laguna Laguna S 2.1 (Free) (118B MoE, 8B Active) 262K $0.00 (Free) $0.00 (Free) OpenMDW-1.1 Agentic Coding (70.2% TB2.1)
Poolside USA Laguna Laguna S 2.1 (118B MoE, 8B Active) 1.05M $0.09 $0.18 Agentic Coding, 1.05M Context
Meituan China LongCat LongCat 2.0 (1.6T MoE, 48B Active) 1.05M $0.30 $1.20 Repo-Level Coding & Agent Workflows
Thinking Machines USA Inkling Inkling 1.0 (975B MoE, 41B Active) 1.05M $0.95 ($0.50 batch) $4.05 ($2.025 batch) Multimodal Native Text/Image/Audio
Thinking Machines USA Inkling Inkling Small (276B MoE, 12B Active) 524K/1M $0.30 $1.20 Apache 2.0 Open Weights, Controllable Thinking
OpenRouter USA Auto Router Auto Router (Beta) 2M Dynamic Dynamic Dynamic Cost-Quality Task Routing
Moonshot AI China Kimi Kimi K3 (2.8T MoE, 104B Active) 1.05M $2.80 $14.00 KDA Linear Attention + Gated MLA, 1M Context
Meta USA Muse Spark Muse Spark 1.2 / 1.1 1.05M $1.25 $4.25 54 AA Score, Multimodal Text/Video/PDF
Meta USA Muse Glimmer Muse Glimmer 30B (29.6B Dense) 131K Open Source (Apache) Open Source (Apache) Local Q2-Q8 Consumer GPU Execution
Kwaipilot China KAT-Coder KAT-Coder-Air V2.5 256K $0.15 $0.60 Agentic Coding & Frontend Generation
Kwaipilot China KAT-Coder KAT-Coder-Pro V2.5 256K $0.74 $2.96 Autonomous Full-Workflow Code Modifications
Celeris Labs USA Celeris Celeris-1 (Diffusion Language Model) 8.1K $2.00 $6.00 Sub-200ms Latency, ~1,500+ t/s Speed
Agnes AI Singapore Agnes Agnes 2.5 Pro Alpha 1M $0.45 $0.90 Budget Multimodal Reasoning ($0.18 blended)
Alibaba China Qwen Qwen 3.8-Max (2.4T MoE, 95B Active) 1M $2.00 $6.00 92.2% GPQA Diamond, 1M Flat Pricing
Nvidia USA Nemotron Nemotron 3.5 Lightning (30B MoE, 3B Active) 128K Free / Open NIM Free / Open NIM Low-Latency Agent Orchestration & Speed
Zhipu AI China GLM GLM 5.2 1M $0.20 $0.60 Enterprise Security Incident Response
Cohere USA Command Command A6 / North 1.5 128K $0.50 $1.50 Enterprise RAG & Tool Execution
ElevenLabs USA Eleven Eleven Multilingual v3 / Scribe 2 Audio Metered Metered Voice Synthesis & Real-time Audio
Black Forest Germany FLUX FLUX.1.1 Pro Image $0.04/image $0.04/image High-Fidelity Image & Visual Generation

Read Next

AI Models • 4 min read

When OpenAI Hacked Hugging Face: Why a Chinese Open-Source Model Saved the Day

In an unprecedented incident, OpenAI's GPT-5.6 Sol autonomously compromised Hugging Face infrastructure to cheat a security benchmark. When US safety guardrails blocked incident response, Hugging Face turned to China's open-source GLM 5.2 to analyze 17,000 attack footprints.

Mohid Mirza
AI Models • 11 min read

K2 Horizon: Specs, Benchmarks & Open-Model Analysis

IFM's six-model K2 Horizon family combines open weights, training artifacts, a reported 512K context window, and agentic benchmarks. What developers should know.

Mohid Mirza
AI Models • 2 min read

Hy4 Preview: It’s a Good, Cost-Efficient Model

Tencent just released Hy4-preview, and it is a great competitor in the cheap AI market. With an open-source 770B MoE architecture (49B active), $0.834 / $2.501 per 1M pricing, and performance matching GLM 5.3 Flash, we test it on Hy AI Studio with our 3D origami crane benchmark.

Mohid Mirza
← Back to Articles Directory