American/Western Models:

Closed Frontier Models:

OpenAI – ChatGPT

• • Valuation:    $852B
  • Owned By:    Sam Altman, OpenAI Foundatoin, Microsoft, Employees & others (Nvidia, Bezos, SoftBank, etc)
• Conversational, mimics human dialogue & banter
• Human feedback trains it to converse & follow complex instruction
• General Purpose: Write, Debug, & translate easily
• Customize specialized mini-chatbots
• Trained on public internet texts, books, articles, code
• Can process & generate all multimedia, process math and can search web

 

Google – Gemini

• • Valuation: $4T  (Alphabet Inc)
  • Owned By:   Alphabet Inc, DeepMind, Google
• Natively multi-modal: reads & writes several media types
• Natively understands video & media, can process & generate all media
• Can process very large amounts of data at once, in a single prompt
• Fully integrates into the entire Google Ecosystem
• Trained on Webpages, books, articles, many languages, programming codes, audio clips, video transcripts & human feedback

 

Anthropic – Claude

• • Valuation:  $965B
  • Owned By:    Independantly owned by founders, developers & employees
• More constitutional, thoughtful -aims to be helpful, honest & harmless
• Less halucinatory, more candor, very transparent
• Will not give incorect info with confidence, will admit if answer is unkown
• Uses more human-like writing style instead of structured bullet points
• Very large context window, can process full volumes of text, books, video at once
• Can do technical analysis, debugging and software architecture
• Opus/Sonnet, etc

 

Xai – Grok

• • Valuation:   $230B – $1.25T
  • Owned By:   Elon Musk, xAI, SpaceX
• Feb by live streaming public posts & data from X (Twitter) to help understand current events, slang, etc.
• Scaled using heavy reinforcment learning that is focused on per-token intelligence & multi-step agentic problem solving
• Has access to real-time knowledge from live breaking news, trends & discussions from X (Twitter)
• Trained on massixe mix of curated web data, code, public real-time conversation data from X (Twitter)
• Has Real-Time data access, multi-modal image & video generation
• Advanced logical reasoning modes and witty, direct personality
• Trained on thousands of specialized accelerators of massive text corpora
• Focuses on data filtering, de-duplication, quality scoring and technical multi-step software tasks
• Has a Helpful Assistant mode as well as a Fun mode that uses humour, wit & sarcasm
• Gives visible step-by-step logic to break down complex math, science and coding problems
• Grok Imagine gives document summarization, image understanding and generative media creation
• Built right into the X platform (Twitter) & mobile apps as well as a standalone app access & API

 

MidJourney

• • Valuation:   $7B – $12B
  • Owned By:   Privately Owned – David Holz
• Highly detailed & stunning Text-to-Image Generator
• Began in Discord & expanded to become its own interface
• Can input an image for use as a theme or reference
• Control Aspect Ratios, styles, artistic flair & other creative deviations
• Trained on public data, 3rd party & user data, Mid-journey data & LLMs

 

Mistral AI

• • Valuation:   $23B
  • Owned By:   3 Founders -former DeepMind & Meta CEO’s & investors
• Based in Paris, France, they bridge a gap between Closed models and Open Weighted models
• Pioneered MoE Architecture -“Mixture of Experts” (a prompt triggers a ‘pathway of parameters’ and may require only related processing pathways rather than the entire capability of the AI). There may be 670B total parameters available, but only 40B parameters (Experts) are activated. This saves computational cost & latency, and delivers Frontier-level reasoning.
• Supports massive context windows allowing 256K tokens to deliver multi-document RAG (Retrieval Augmented Generation), Large Database processing and long form analytical reports
• Can deploy autonomous agents through Mistral AI Studio
• Trained on massive amount of public internet data, code repositories spanning over 80 programming languages (logic, syntax, architectural), and synthetic data generation pipelines to fine tune & optimize reasoning layers

 

---

 

 

Open Weighted Models:

Perplexity

• • Valuation:   $21B
  • Owned By:   Privately owned by former OpenAI, Google AI and Microsoft AI guys, Meta, Quora, Databricks, Investors (Bezos, NVidia, etc)
• Each querry starts an active web search, not memory -not memory, source links
• Can select between models depending on needs (GPT, Claude, etc)
• Focus Mode restricts searches to academic, social or financial papers
• Deep Research uses parallel searches autonomously for full reports
• Fine Tunes external models (GPT, Claude, Gemini) Uses RAG processing
• Rag -Retrieval Augmented Generation -Uses live content to process responses (Bing, Google, MS Graph data & Live web data)

 

MicroSoft – Co-Pilot

• • Valuation:  Unknown
  • Owned By: Microsoft  – Built on Open AI
• Conversational & Generative
• Can process text, voice, images & create visuals from descriptions
• Embedded in all MS-365 apps (Office, Teams, etc)
• Dual-Mode grounding – Web mode / Work mode
• Web Mode -retrieves public data via Bing
• Work Mode -retrieves organizatioal data via MS Graphs
• Agentic -uses autonomous companions customized via MS Co-Pilot Studio
• Create strict boundaries from MS-365 Permissions & rules
• Uses OpenAI’s  LLM’s (GPT) & MS-Promethius framework (Bing)
• Uses RAG (Retrieval Augmented Generation (MS-Graph & Live Web data)

GitHub – Co-Pilot

• • Valuation:  Unknown
  • Owned By:   Microsoft, GitHub (Uses OpenAI)
• Also controlled by MicroSoft but is separate from preious Co-Pilot
• Multi-Modal -Image, text, video, code
• Selectable LLM  (GPT, Gemini, Claude)
• Assists with real-time coding, integrates into code editors
• Chat Mode or Agent Mode -can communicate and execute
• Context-Aware Auto Complete based on your open tabs & cursor positions
• Trained on GPT -public code repositories, open source code (GitHub, illustrations and conceptualizations), Natural Language texts, billions of lines of source code from public repositories

 

 

Eleven Labs

• • Valuation:   $11B – $22B
  • Owned By:   2 Co-Founders in Poland, Minority Investors (Banks in Poland)
• Primarily Text-to-Speech, voice cloning, translation
• Perceives emotional depth context
• Trained on 100K hrs of multi-speaker dataspoken word & audio books

 

Grammarly

• • Valuation:   $13B (min)
  • Owned By:   SuperHuman Platform Inc, Investors in Ukraine, Blackrock
• Complete writing correctness, clarity & readability enhancements
• Can set target goals based on audience & more
• Use dedicated Agents lie Humanizer and Reader Reactions
• Plaigerism & AI Detection Prevention
• Enterprise and Team Customization
• Integrates into browser extensions, apps and editors
• Trained on LLM’s, NLP’s (Natural Language Processing), rule based algorythms and machine learning

 

Notion AI

• • Valuation:   $11B
  • Owned By:Co-Founders & a few Investors (American)
• LLM does NOT train on your personal/workspace data by default
• Workspace-wide search of connected apps
• Auto Content Generation and editing
• Uses autonomous multi-step Agents
• No User Training -Uses external models (GPT, Gemini, Claude) to bridge top LLM’s rather than be a proprietary standalone model
• Draft documents, fix grammar, summarize, translate, re-write (change the tone of the output)
• Create databases, content, calendars, and tables of data

 

Zapier

• • Valuation:   $5B
  • Owned By:   Zapier Inc Co-Founders
• An orchestration layer for frontier models ((GPT, Claude, Gemini)
• Connects the LLM to over 9K apps
• Can troubleshoot workflows using language
• Autonomous systems & chatbots can execute tasks in specific apps
• Can access & use specific data sources
• Translates instructions to external AI’s, so that the AI can invoke Zapier app actions

 

Jasper

• • Valuation:   $1.2B – $1.5B
  • Owned By:   Co-Founders, Execs & Venture Capital Firms
• Uses all the top LLM’s (GPT, CLaude, Gemini)
• Specifically designed & engineered for marketing projects
• Jasper IQ -shared intelligence layer with brand, product and target audience in mind
• Combines top AI models with SEO & Fact Checking integrations
• Agentic -create agents that specialize in marketing roles
• Trained on a vast amount of public internet text via LLM’s & multimode engines as well as uploaded brand assets (company data)

 

Azure

• • Valuation:   $135B
  • Owned By:   Microsoft Cloud Based, Uses OpenAI
• A MicroSoft platform of cloud-based tools that use external Frontier LLM’s and accesses user-provided data to generate a private model via MicroSoft Foundary
• Organizations upload proprietary datasets, documents, images & audio
• Strict governance, your data stays private
• Ready to use API’s for vision, speech, language & translation
• Unified hub to discover, test, build & deploy generative apps & agents

 

---

 

Chinese Models:

 

Moonshot AI  – Kimi K3

• • Valuation:   $30B – $50B
  • Owned By:
• Tailored specifically for Long-Horizon Autonomy, terminal execution and software developnment, Kimi 3 outperforms many western models
• Supports Enterprise customization and academic research
• Very large 1 million token context window with native visual/video translation
• 2.8trillion total parameters, 896 experts, using only 16 active at a time (104B active parameters) per token for high inference efficiency
• Trained on very diverse multi-modal cirriculum of 45 trillion tokens of text, image, audio/video and reinforcement learning

 

DeepSeek

• • Valuation:   $50B – $70B
  • Owned By:
• Operates on 671B total parameters, acts on only 37B per token to save power, uses multi-head latent attention
• Uses 8-bit floating point training to reduce memory movement & hardware demands
• Shows internal train of thought (step-by-step) before outputting answers
• Open licensing, developers can run & distill smaller versions locally
• Developed & trained at a fraction of the cost of Western models
• Trained on trillions of diverse high-quality public & licensed text tokens covering general, math, coding, multi-lingual text, as well as Reinforcement Learning (Group Relative Policy Optomizing) to teach how to think as well as self-verify

 

 

AliBaba – Qwen

• • Valuation:   $275B
  • Owned By:   Ali Baba
• Open Weighted/Open Source variants and larger proprietary cloud variations
• 18 – 36 trillion tokens depending on model tier level
• Rigorous automated safety filters, quality cleaning and human alignment training
• Multi-modal:  Qwen processes text, vision, audio/video streams
• Multi-lingual, math, coding, science, law, medicine
•  Cheaper and nearly as good as western models
• Trained on massive datasets -10’s of trillions of text tokens, code repositories, academic papers, multi-modal inputs (images, audio/video)

 

GLM – Z.ai

• • Valuation:   $72B – $128B
  • Owned By:   GLM (Originally Zhipu AI, Knowledge Atlas Tech)
• Family of Open-Weight LLM’s trained on massive multi-modal text & code datasets
• Released under Open Source/Open-Weight MIT-style licenses on platforms like HuggingFace
• Massive context window -1 million tokens of context input window, large output generation
• MoE Architecture (Mix of Experts) and domestic hardware stacks (like Huawei Ascend)
• Long-Horizon Agents to execute complex multi-step autonomous workflows & coding tasks continuously for hours, without losing coherence
• Competitive performance in coding & reasoning at a fraction of the cost of western closed models
• Powers specialized developer environments & coding assistants (like Z-Code)
• Advanced training recipes and sparse attention mechanisms with massive parameter scales (753B parameters in newer models)