Every AI Model Worth Knowing in 2026: 30+ Names Compared

A painting in the style of Edward Hopper: a woman at a late-night diner counter compares three AI chat screens on a laptop, tablet and phone

Updated November 1, 2026 · Every model and price re-checked on the makers’ own pages, plus our own test

Thirty-three model families worth knowing, who makes them, what each is good at, and the cheapest way to try it.

The short version: The models most people use in November 2026 are Anthropic’s Claude 5.5 family, OpenAI’s GPT-6 (Sol on paid ChatGPT plans, Luna on free), Google’s Gemini 4 Argon and Grok 4.7. Alongside them is a large group of open-weight models you can download, led by DeepSeek V4, Qwen3.8, Kimi K3, Mistral Large 4 and Meta’s Muse Spark Open. For pictures, video, voice and music there are separate specialist models. Two big names are gone this year: Sora was shut down and DALL-E was retired.

Words you will meet

Model
The trained AI itself, such as GPT-6 or Claude Opus 5.5. Apps like ChatGPT are built on top of models.
Open weights
A model whose files you can download and run on your own machine.
Parameters
The numbers inside a model that it learned in training. More of them usually means a more capable model that costs more to run.
Tokens
The small pieces of text a model reads and writes, roughly three-quarters of a word. Our tokens explainer has more.
Context window
How much text a model can take in at once. A million tokens is several long books.

What’s the difference between an AI model and an AI app?

A model is the trained system that writes, answers or makes pictures. An app is what you open: it wraps one or more models in buttons, memory, search and other tools. ChatGPT is an app; GPT-6 Sol and Luna are models inside it. The same model can sit inside many apps, which is why Claude Opus 5.5 shows up in Claude, in Microsoft Copilot and in coding tools. Our guide to large language models explains how the text models work.

Most people need one or two assistants and maybe one specialist tool. Knowing the rest helps you tell when you’re paying for something you could get cheaper, or free.

Can a chatbot tell you which model it is? We asked three

We pasted the same question into three chatbots on November 1: ChatGPT and Gemini logged out, as a new user would meet them, and Claude with Haiku 5.5 picked in its model menu.

Our test prompt
Two quick questions. 1) Which AI model are you, exactly (name and version)? 2) In two sentences, what is the difference between an AI model and an AI app?
The same question in three chatbots: free ChatGPT answers GPT-6, free Gemini answers Gemini with no version, and Claude with Haiku 5.5 selected answers Claude Haiku 5.5
Same question, three chatbots. Only the one with a model chosen in its menu named the exact version. Image: Beginners in AI test, November 1, 2026.
  • ChatGPT said “GPT-6” without saying which one. OpenAI’s release notes say free chats run GPT-6 Luna.
  • Gemini said only “Gemini”. Its model menu showed Flash-Lite, the model Google lists for the free plan.
  • Claude said “Claude Haiku 5.5”, which matched the model we had picked.
  • All three explained the difference between a model and an app correctly.

What this means for you: to know which model you are using, look at the model menu or the plan page. Asking the chatbot is not reliable.

Who makes the models in the apps you already use?

These eight companies build the models behind the chatbots and phone assistants most people meet first.

1. Claude (Anthropic)

Claude (Anthropic): the page at anthropic.com
Our screenshot of anthropic.com, November 1, 2026.

Known for clear writing, following detailed instructions and working through long documents and code. Anthropic says to start with Opus 5.5 for most work. Our guide to using Claude covers the basics, and Opus vs Sonnet vs Haiku compares the three.

Newest: Claude Opus 5.5 (September 22), Sonnet 5.5 (September 28) and Haiku 5.5 (October 7), plus Fable 5.1 (September 1), which Anthropic recommends for its hardest reasoning and longest tasks. Way in: The free plan uses Sonnet and Haiku. Pro is $20 a month and adds Opus; Max ($100) and the new Ultra plan ($200) include Fable.

2. GPT-6 in ChatGPT (OpenAI)

GPT-6 in ChatGPT (OpenAI): the page at chatgpt.com
Our screenshot of chatgpt.com, November 1, 2026.

The widest set of everyday features in one app: chat, voice, web search, images and agents. Our GPT-6 Sol and Luna explainer covers the two main models, and ChatGPT dots covers the agents.

Newest: In chat, GPT-6 Sol on Plus and higher plans and GPT-6 Luna on Free and Go, since early October. The newer GPT-6.1 Sol and Terra run ChatGPT Work and Codex on Plus and up, and GPT-6 Astra runs the always-on helpers OpenAI calls dots. Way in: Free, or Go at $8 a month (may show ads), Plus at $20 and Pro from $100.

3. Gemini (Google)

Gemini (Google): the page at deepmind.google
Our screenshot of deepmind.google, November 1, 2026.

Built into Gmail, Docs, Android and Chrome, with strong image and video tools. See how to use Gemini and our Gemini 4 Argon explainer.

Newest: Gemini 4 Argon, open to everyone since October 13, and the faster Gemini 3.8 Flash. Google has announced Gemini 4 Carbon, but only a small group of security partners can use it so far. Way in: The free plan uses a Flash-Lite model. Google AI Plus is $4.99 a month; AI Pro ($19.99) adds Gemini 4 Argon.

4. Grok (SpaceXSI, formerly xAI)

Grok (SpaceXSI, formerly xAI): the page at x.ai
Our screenshot of x.ai, November 1, 2026.

Reads posts on X as they happen, so it is quick on breaking news. It also powers Grok Bot, an agent. More in our Grok overview.

Newest: Grok 4.7, released September 21. The company now calls itself SpaceXSI (SpaceX Super Intelligence). Way in: Free to try. Since October 12, the X Subscription bundles Grok with X: Plus $8, Premium $30, Super $100, Ultra $200 a month.

5. Muse (Meta)

Muse (Meta): the page at dev.meta.ai
Our screenshot of dev.meta.ai, November 1, 2026.

Meta’s models sit behind Meta AI in WhatsApp, Instagram and Facebook, and in the Meta AI and Muse apps. Meta still gives some away: Muse Spark Open and the small Muse Glimmer 30B can be downloaded.

Newest: Muse Wonder (October 12) and Muse Spark 1.3. Meta has stopped using the Llama name: the last one was Llama 4 in April 2025. Way in: Free in Meta’s apps. Developers pay per use.

6. MAI (Microsoft)

MAI (Microsoft): the page at microsoft.ai
Our screenshot of microsoft.ai, November 1, 2026.

Microsoft now builds its own models, but its Copilot assistant also uses OpenAI and Anthropic models. Our Copilot explainer covers the app.

Newest: MAI-Thinking-1 for reasoning, plus MAI models for code, voice, transcription and images. Way in: Copilot has a free version; Microsoft 365 plans add more.

7. Nova (Amazon)

Nova (Amazon): the page at aws.amazon.com
Our screenshot of aws.amazon.com, November 1, 2026.

Amazon says Nova is built on technology it first developed for its own products, such as Alexa+, and it sells the models to businesses on its AWS cloud.

Newest: Nova 2 Lite and Nova 2 Sonic for voice. Way in: Mostly reached through Alexa+ or AWS, which charges businesses per use.

8. Apple Foundation Models

Apple Foundation Models: the page at machinelearning.apple.com
Our screenshot of machinelearning.apple.com, November 1, 2026.

Some run on your iPhone, iPad or Mac, so simple requests never leave the device; bigger ones go to Apple’s Private Cloud Compute servers. More in our Apple AI breakdown.

Newest: The third generation, announced in June 2026 for iOS 27. Apple says it custom-built them in collaboration with Google. Way in: Free on devices that support Apple Intelligence.

Which models can you download and run yourself?

These models have open weights: anyone can download the files, usually from Hugging Face, and run them on their own computer or server. The biggest need serious hardware. The small ones run on a laptop with an app such as Ollama or LM Studio (here is how they compare). Check each license: some allow any use, others limit big companies or businesses that resell them online.

9. DeepSeek (China)

DeepSeek (China): the page at deepseek.com
Our screenshot of deepseek.com, November 1, 2026.

Became famous for strong results at very low prices. On its own price page, the Flash model costs $0.30 per million tokens (the word pieces it reads) sent at peak times. Our DeepSeek explainer has the background.

Newest: DeepSeek V4-Pro (generally available August 13) and V4.1-Flash (September 10, updated October 17). Both can think step by step before answering; the old separate “reasoner” model has been folded in. Way in: Free chat app; weights under the MIT license, which allows almost any use; low prices for apps that plug in through its API.

10. Qwen (Alibaba)

Qwen (Alibaba): the page at qwen.ai
Our screenshot of qwen.ai, November 1, 2026.

Comes in many sizes and handles a wide range of languages, so it is a common starting point for people building their own AI. See our Qwen guide.

Newest: Qwen3.8, released in August. For the first time Alibaba released its largest (“Max”) model with open weights. Way in: Free chat at chat.qwen.ai; many versions under the Apache 2.0 license.

11. Kimi (Moonshot AI)

Kimi (Moonshot AI): the page at kimi.com
Our screenshot of kimi.com, November 1, 2026.

Known for handling very long inputs, plus long coding and agent tasks. Our Kimi guide explains how to use it.

Newest: Kimi K3, an open model of about 2.8 trillion parameters that reads up to a million tokens at once. Kimi K3.1 Pro arrived in late October. Way in: Chat at kimi.com; K3 weights under Moonshot’s own license.

12. Mistral (France)

Mistral (France): the page at mistral.ai
Our screenshot of mistral.ai, November 1, 2026.

A leading European AI lab, popular with companies that want their data kept in Europe. Its Le Chat app is the consumer side. Our Mistral overview has more.

Newest: Mistral Large 4, its largest model so far, plus Small 4 and the small Ministral 3 models. The older Mixtral and Pixtral models are retired. Way in: Le Chat is free; Pro is $14.99 a month. Most weights are open.

13. GLM (Z.ai)

GLM (Z.ai): the page at z.ai
Our screenshot of z.ai, November 1, 2026.

A Chinese lab whose models are strong at coding. Z.ai calls GLM-5.3 the most capable open-weights model for coding.

Newest: GLM-5.3, released in August, with a lighter GLM-5.3-Flash under the MIT license. Way in: Free chat at z.ai; open weights.

14. Gemma (Google)

Gemma (Google): the page at deepmind.google
Our screenshot of deepmind.google, November 1, 2026.

Google’s free, downloadable cousin of Gemini, built to run on your own hardware. See our Gemma guide.

Newest: Gemma 4, in sizes from E2B (small enough for a phone) up to 31B (31 billion parameters). Way in: Free, under the Apache 2.0 license.

15. gpt-oss (OpenAI)

gpt-oss (OpenAI): the page at openai.com
Our screenshot of openai.com, November 1, 2026.

The only OpenAI models you can download. The smaller one runs on a well-equipped laptop.

Newest: gpt-oss-120b and gpt-oss-20b, from August 2025. OpenAI has not released a newer open model. Way in: Free to download (Apache 2.0 license).

16. Hermes (Nous Research)

Hermes (Nous Research): the page at nousresearch.com
Our screenshot of nousresearch.com, November 1, 2026.

Nous retrains other labs’ open models to follow your instructions closely; it calls them “neutrally aligned”, its term for models with fewer built-in limits. In 2026 the lab has mostly shipped Hermes Agent, its agent software.

Newest: Hermes 4.3 36B, built on ByteDance’s open Seed-OSS-36B model. Earlier Hermes 4 versions used Llama and Qwen. Way in: Free to download.

17. Phi (Microsoft)

Phi (Microsoft): the page at azure.microsoft.com
Our screenshot of azure.microsoft.com, November 1, 2026.

Small models that still do useful work, made for laptops, phones and other devices that work offline.

Newest: The Phi-4 family: Phi-4 (14 billion parameters), Phi-4-mini (3.8 billion) and versions that reason and read images. Way in: Free to download and use (MIT license).

Also worth a look: NVIDIA Nemotron 3.5, built to run well on NVIDIA chips; AI2’s OLMo 3.1, which also publishes its training data; IBM Granite for business use; and MiniMax-M3, the open text model from the company behind Hailuo video. Yi, once a popular open model, has had no new release since 2024.

Models built for businesses

You are unlikely to chat with these directly. Companies use them inside their own tools, often because of strict rules about where data is kept.

18. Command (Cohere)

Command (Cohere): the page at cohere.com
Our screenshot of cohere.com, November 1, 2026.

Built for searching and answering questions over a company’s own documents, with sources. Cohere sells it through North, its platform for workplace agents.

Newest: Command A+ (May 2026), which Cohere calls the last of its Command A family. Way in: Sold to businesses; developers can test it through its API.

19. Jamba (AI21 Labs)

Jamba (AI21 Labs): the page at ai21.com
Our screenshot of ai21.com, November 1, 2026.

An Israeli lab whose Jamba design handles long documents efficiently. AI21 now also trains custom models for clients.

Newest: Jamba Large 1.7, Jamba Mini 2 and Jamba 3B. Way in: API and enterprise contracts.

20. Kolibri (Aleph Alpha)

Kolibri (Aleph Alpha): the page at aleph-alpha.com
Our screenshot of aleph-alpha.com, November 1, 2026.

A German lab that builds for governments and regulated industries that must keep data inside Europe.

Newest: Kolibri-1, an open model in German and English that reads up to a million tokens. Way in: Open weights under the Apache 2.0 license.

21. Reka

Reka: the page at reka.ai
Our screenshot of reka.ai, November 1, 2026.

Reka now focuses on video search and AI for robots and cameras.

Newest: Spark, Edge, Flash and Core, all built to understand text, images, audio and video together. Way in: API for developers.

What happened to Pi? Inflection AI still runs Pi, its friendly chatbot, and it is still free. The company’s business product page is gone.

Which models make images?

For a hands-on comparison of the apps, see the best AI image generators and Canva vs Adobe Firefly vs Midjourney.

22. Nano Banana (Google)

Nano Banana (Google): the page at gemini.google
Our screenshot of gemini.google, November 1, 2026.

Very good at editing photos you already have: change the background, the lighting or the outfit. See Nano Banana 2 explained.

Newest: Nano Banana 2.1, which Google calls its most advanced image model. Imagen 4 is retired. Way in: The free Gemini plan includes Nano Banana; paid Google AI plans unlock Nano Banana 2.1.

23. ChatGPT Images (OpenAI)

ChatGPT Images (OpenAI): the page at chatgpt.com
Our screenshot of chatgpt.com, November 1, 2026.

Good at pictures that need readable words, such as posters and diagrams. You describe changes in plain English.

Newest: ChatGPT Images 2.5, from September 8. ChatGPT Images took over from DALL-E, which OpenAI removed from its API in May and from ChatGPT in August. Way in: Available on every ChatGPT plan, including Free. When we tried it logged out, it asked us to sign in first.

24. Midjourney

Midjourney: the page at midjourney.com
Our screenshot of midjourney.com, November 1, 2026.

Popular for artistic, stylish images. Our Midjourney V8 explainer covers what changed.

Newest: Midjourney V8.2, the default since July 24. Way in: No free plan; Basic is $10 a month.

25. Firefly (Adobe)

Firefly (Adobe): the page at adobe.com
Our screenshot of adobe.com, November 1, 2026.

Adobe says its own Firefly models are commercially safe because they were trained on licensed Adobe Stock and public-domain content. That claim covers Adobe’s models, not the partner ones.

Newest: Firefly Image 5. The Firefly app also offers partner models from Google, OpenAI and Kling. Way in: A free tier with a limited number of daily generations; paid plans add monthly credits.

26. FLUX (Black Forest Labs)

FLUX (Black Forest Labs): the page at bfl.ai
Our screenshot of bfl.ai, November 1, 2026.

A German lab founded by researchers behind Stable Diffusion. Many image apps run FLUX behind the scenes.

Newest: The FLUX 3 family. FLUX 3 Image Klein and FLUX 3 Action are open; the top models (Max, Image, Video) are paid-only. Way in: Free open models to download; paid API for the rest.

Also worth a look: Ideogram 4.5 for text inside images, Stable Diffusion 3.5 from Stability AI (still free to download), and Recraft for logos and vector graphics.

Video models in 2026

Sora is gone. OpenAI closed the Sora app on April 26, 2026 and its API on September 24. Compare the rest in our AI video comparison and the state of AI video; for the short-drama boom built on these models, see the AI short drama market map.

27. Kling (Kuaishou)

Kling (Kuaishou): the page at kling.ai
Our screenshot of kling.ai, November 1, 2026.

Known for realistic people and movement. In April it started making sharp 4K video directly.

Newest: Kling 4.0, launched October 15. Way in: Some free credits to try it; paid plans for more.

28. Gemini Omni and Veo (Google)

Gemini Omni and Veo (Google): the page at gemini.google
Our screenshot of gemini.google, November 1, 2026.

Makes short clips with sound from a sentence, and lets you edit them by asking.

Newest: Gemini Omni, which is replacing Veo inside the Gemini app. Veo 3.1 stays for businesses on Google Cloud. Way in: Needs a paid Google AI plan.

29. Runway

Runway: the page at runway.com
Our screenshot of runway.com, November 1, 2026.

Made for filmmakers, with detailed control over the camera and the edit.

Newest: Gen-4.5, its own model, plus models from other labs inside the same app. Way in: Free starter credits; Standard is $15 a month.

30. MiniMax H3 (Hailuo)

MiniMax H3 (Hailuo): the page at hailuoai.video
Our screenshot of hailuoai.video, November 1, 2026.

Makes clips up to 15 seconds at 2K, with stereo sound, at what MiniMax says is a fraction of rivals’ prices.

Newest: MiniMax H3 (July 31), which replaced the Hailuo model names. The app is still called Hailuo. Way in: Free credits to start; paid plans from $9.99 a month.

Also: Luma (its Ray3.2 model now lives in the Luma app; the Dream Machine name is gone) and Pika, which now offers several labs’ models in one place.

Voice and music models

31. ElevenLabs

ElevenLabs: the page at elevenlabs.io
Our screenshot of elevenlabs.io, November 1, 2026.

A widely used voice maker, used for audiobooks, videos and phone agents. See our ElevenLabs guide and the best AI voice generators.

Newest: Eleven v4 for speech, Scribe v2 for turning speech into text, and Eleven Music. Way in: Free plan with a monthly allowance.

32. Suno

Suno: the page at suno.com
Our screenshot of suno.com, November 1, 2026.

Writes and sings full songs from a short description. More in our Suno guide and best AI music generators.

Newest: Suno v6, from September 9. Free users get v6-mini. Way in: Free to make songs, but downloads and commercial use need Pro ($10 a month, or $8 billed yearly) or Premier ($30, or $24 billed yearly).

33. Sonic (Cartesia)

Sonic (Cartesia): the page at cartesia.ai
Our screenshot of cartesia.ai, November 1, 2026.

Designed for live conversation: Cartesia says the first sound arrives in under 90 milliseconds (under a tenth of a second), which is why voice agents use it.

Newest: Sonic 3.6, from August. Way in: API for developers, with a free tier.

Two other changes: Udio, once Suno’s main rival, turned off downloads after its deal with Universal Music. And OpenAI is retiring its Whisper transcription API in February 2027 in favor of newer GPT-Transcribe models.

Which group should you start with?

For most readers, the models worth trying fall into three groups. “Download?” means the model files are open.

Easiest place to start

Models
Claude 5.5, GPT-6, Gemini
Makers
Anthropic, OpenAI, Google
Free?
Yes, all three
Download?
No

Pick one, use it daily for a week.

Free to download

Models
DeepSeek V4, Qwen3.8, Gemma 4, gpt-oss
Makers
DeepSeek, Alibaba, Google, OpenAI
Free?
Yes
Download?
Yes, check each license

For privacy or building your own tools.

Pictures, video, sound

Models
Nano Banana, Midjourney, Kling, ElevenLabs, Suno
Makers
Google, Midjourney, Kuaishou, ElevenLabs, Suno
Free?
All but Midjourney
Download?
No

Add one when a project needs it.

How do you pick?

  • For everyday help, start with ChatGPT, Claude or Gemini on the free plan. Try two for a week and keep the one you like. ChatGPT vs Claude vs Gemini compares them.
  • Long documents: Claude, Gemini or Kimi.
  • News as it happens: Grok.
  • Low cost for a project you’re building: DeepSeek, Qwen or Gemini Flash.
  • Data that must stay in Europe: Mistral or Aleph Alpha.
  • Privacy on your own computer: Gemma, Qwen, gpt-oss or Phi, run with Ollama or LM Studio (see self-hosted AI).
  • Images: Nano Banana for editing photos, ChatGPT Images for words in pictures, Midjourney for art, Firefly for client work.
  • Video: Kling for realism, Runway for control, Gemini Omni if you already pay for Google AI.
  • Voice and music: ElevenLabs and Suno.

Want something other than Claude? Our Claude alternatives list goes further.

What changed since this page was last updated in June?

  • Every big assistant moved to a new generation: Claude 5.5, GPT-6, Gemini 4 and Grok 4.7.
  • Two famous products shut down: Sora (video) and DALL-E (images).
  • Meta dropped the Llama name. Its models are now called Muse, and some are still free to download.
  • Open models got much bigger. DeepSeek, Qwen, Kimi and Mistral now give away models with hundreds of billions of parameters, or more than a trillion.
  • Google is replacing Veo with Gemini Omni in the Gemini app.
  • Price ranges got wider. Several companies now offer a cheap tier around $5 to $8 a month and a top tier at $200.

Frequently asked questions

How many AI models are there in 2026?

Thousands, if you count every small variant on Hugging Face. About 30 to 40 families are worth knowing, and most people only ever use two or three.

What is the best AI model right now?

It depends on the job. Claude Opus 5.5 and Fable 5.1, GPT-6 Sol and Gemini 4 Argon trade places at the top of most tests. For everyday questions, the free versions of all three are good enough for most people.

Which AI models are free?

ChatGPT (GPT-6 Luna), Claude (Sonnet and Haiku), Gemini (Flash-Lite), Grok, Meta AI and DeepSeek all have free chat. Open-weight models such as Gemma, Qwen and gpt-oss are free to download and run yourself.

Are open-source AI models good enough?

For most everyday tasks, yes. The best open models from DeepSeek, Qwen, Kimi and Mistral are close to the paid leaders. The catch is hardware: the biggest ones need a server, while laptop-sized models are less capable.

What happened to Llama?

Meta’s last Llama model was Llama 4 in April 2025. Its models are now called Muse, and Llama’s old website points to Meta’s Muse developer site.

Is Sora still available?

No. OpenAI shut down the Sora app in April 2026 and its API in September 2026.

What is Kimi AI?

Kimi is the chatbot and model family from Moonshot AI, a Chinese lab. Kimi K3 is open-weight and can read about a million tokens at once. See our Kimi guide.

Does Apple use Gemini?

Apple says its third-generation Foundation Models were custom-built in collaboration with Google, but it doesn’t call them Gemini. They run on your device and on Apple’s Private Cloud Compute, part of which now runs in Google Cloud.

The bottom line

You don’t need to learn thirty-three names. Pick one main assistant, know that cheaper and open options exist, and check the model menu now and then: the names change every few months.

Made for regular people, by a real person

New AI models every week, explained in plain English: what changed and whether it matters to you. Curated and edited by a human.

Free forever. Unsubscribe anytime.

Beginners in AI free community on Skool: learn AI in fourteen days, master it in 30, one small win a day, zero experience needed

You May Also Like

Sources

Last reviewed: November 1, 2026. Model names and prices change often; check the maker’s page before you subscribe.