Types of AI in 2026: 8 Kinds of Generative AI and the Top Model in Each

Types of AI are easiest to understand when sorted by what they produce, and as of September 2026 the generative AI people use most falls into 8 kinds: text, image, video, voice, music, coding, search and research, and presentations. The company holding first place also differs by field. According to Artificial Analysis as of September 25, 2026, the top text-intelligence model is Anthropic's Claude Opus 5.5, the top image model is OpenAI's GPT Image 2.5 Sunburst, the top video model is Google's Gemini Omni Flash, and the top text-to-speech model is Cartesia's Sonic 3.6. Below, each kind is covered by what it makes and what it is good at, followed by the technical categories people often search for: generative versus discriminative AI, and weak versus strong AI.

Types of AI at a glance: 8 kinds by output

The most practical way to choose among types of AI is by the output you want, not the input. Whether the result is text, an image or sound changes how performance is measured and which companies lead. As of September 2026 the 8 kinds and their representative models are as follows.

Type	What it makes	Representative models and tools (Sept 2026)	Good for
Text (LLM, chatbots)	Writing, answers, summaries	Claude Opus 5.5, GPT-6 Astra	Q&A, drafting, analysis
Image generation	Illustrations, photo-style images	GPT Image 2.5 Sunburst, Nano Banana 2	Thumbnails, ad mockups, illustration
Video generation	Short videos	Gemini Omni Flash, Wan 3.0	Short-form clips, scene drafts
Text-to-speech (TTS)	Human-like voices	Sonic 3.6, Gemini 3.8 Flash TTS	Narration, dubbing, voice prompts
Music generation	Songs, background tracks	Suno	Background music, song sketches
Coding agents	Code, terminal work	Claude Opus 5.5, GPT-6 Astra	Bug fixes, feature work
Search and research	Reports with citations	Gemini Notebook Deep Research	Research, checking sources
Presentations	Slides, infographics	Gemini Notebook Slide Decks	Summaries, deck drafts

The text, image, video, speech and coding rows list the top models in the September 25, 2026 edition of Artificial Analysis data shown on the ASAP leaderboards. Music, search and research, and presentations have no comparable ranking, so those rows list services whose features were confirmed on official product pages.

Text AI (LLM) types and rankings

Text AI is the kind where large language models (LLMs) read and write, and chatbots such as ChatGPT, Gemini and Claude all belong to it. In the September 25, 2026 edition of the Artificial Analysis Intelligence Index, first place is Anthropic's Claude Opus 5.5 at 58 points, followed by Anthropic's Claude Fable 5.1 and OpenAI's GPT-6 Astra at 53 points each.

Text AI splits into two branches. One is closed models that run only on the vendor's servers; the other is open-weight (open-source) models whose weights anyone can download and run. The top open-weight model is Xiaomi's MiMo-V2.6-Pro at 46 points, which also ranks 7th overall. For organizations that cannot send personal data or internal documents outside, open-weight models are the option to consider. Live rankings are kept current on the [LLM intelligence leaderboard](/leaderboard/llm-intelligence/) and the [open-source LLM leaderboard](/leaderboard/open-llm/).

Image and video generation AI types and top models

Image and video generation AI is ranked by blind-arena Elo, where people compare two outputs without knowing their source and vote for the better one. In the September 25, 2026 Artificial Analysis edition, first place in image generation is OpenAI's GPT Image 2.5 Sunburst at Elo 1196, with OpenAI's GPT Image 2.5 Flare (1190) and GPT Image 2 (1171) in second and third.

Video generation looks different. First-place Google Gemini Omni Flash (Elo 1233), second-place Alibaba Wan 3.0 (1229) and third-place Minimax H3 Max from fal (1227) sit within 6 points of one another. The open-weight MiniMax H3 ranks 4th overall at 1220, just 13 points behind the top closed model. By contrast, the top open-weight image model, Ideogram 4.0 (Quality), sits at Elo 1010, more than 180 points behind the image leader. Anyone looking for a model to run for free has wider choice in video than in images. Current rankings are on the [image generation leaderboard](/leaderboard/image-generation/) and the [video generation leaderboard](/leaderboard/video-generation/).

Voice, music, coding and search AI types

Text-to-speech (TTS) AI is the kind that reads text aloud in a human voice, and first place in the September 25, 2026 Artificial Analysis Speech Arena is Cartesia's Sonic 3.6 at Elo 1279. Google's Gemini 3.8 Flash TTS is second at 1265 and Alibaba's Qwen-Audio-3.0-TTS-Plus third at 1259. The ranking is on the [text-to-speech leaderboard](/leaderboard/text-to-speech/).

• Music generation AI: Turns lyrics or a mood into a song. The representative service, Suno, describes its product as a browser-based studio suite for shaping songs. ASAP does not yet have a music leaderboard.
• Coding agents: Go beyond writing code to running commands in a terminal and finishing the task. On Terminal-Bench v4.0, which measures 66 tasks, Claude Opus 5.5 and GPT-6 Astra tie for first at 59.6%.
• Search and research AI: Searches the web or your documents and produces answers with citations. Deep Research in Google's Gemini Notebook (formerly NotebookLM) browses up to hundreds of websites on its own and writes a report, and the same tool's Studio also makes slide decks and infographics, doubling as a presentation AI.

Generative AI vs discriminative AI, weak AI vs strong AI: the differences

Generative AI is AI that creates new outputs, while discriminative AI is AI that identifies what a given piece of data is. Spam filtering, face recognition and defect detection are typical discriminative uses, while the chatbots and image and video generators drawing search volume in 2026 are all generative. Weak versus strong AI is a separate axis based on the range of tasks a system can handle.

Category	Core question	Examples	Status, Sept 2026
Generative AI	Does it create something new?	Chatbots, image, video and voice generation	Center of consumer services
Discriminative AI	Does it identify what something is?	Spam filters, face recognition, defect detection	Still used in search, security and manufacturing
Weak AI	Does it only do tasks within a set scope?	All commercial AI today	The norm
Strong AI	Can it do any intellectual task like a human?	Artificial general intelligence (AGI) debate	No agreed-upon example

The two axes overlap. A chatbot like ChatGPT is both generative and weak AI, because as of September 2026 no AI is agreed to have reached human-level general intelligence, however many tasks it handles.

What the 2026 AI map shows: a different leader in each field

The defining feature of the September 2026 AI map is that no single company holds every field. Looking only at first place in the September 25 Artificial Analysis edition, text belongs to Anthropic, images to OpenAI, video to Google and speech to Cartesia, four different companies in four fields. Even OpenAI, which holds the top 3 image spots, has no model in the video generation top 10.

The implication for users is clear. "Just recommend one AI" has no single right answer; each field needs its own pick. And in fields where the leaders are tightly bunched, such as video generation with the top 3 within 6 points, price and ease of use decide the choice more than rank. Scores shift every month as votes and evaluations accumulate, so the rankings in this article should be read as a September 25 snapshot.

AI recommendations: how to choose the right AI for you

The fastest way to choose an AI is to decide the type first, narrow the candidates to two or three on that type's ranking, and then try them yourself. As of September 2026, the 6 ASAP leaderboards summarize Artificial Analysis data in Korean, so narrowing the field takes only minutes.

1. Decide the output: Write down whether you need text, images, video, a voice or code.
2. Pick 2-3 candidates from the ranking: Look at the top of that leaderboard, and treat models separated by small score gaps as effectively the same tier.
3. Filter by constraints: Check Korean-language quality, free availability, whether your data may leave your organization, and commercial-use terms. If you handle internal documents, add open-weight models to the list.
4. Compare with the same input: Give every candidate the same question or prompt and put the results side by side.
5. Check again in a month: The September 25 edition alone brought changes at or near the top in text, image, video and speech. Look at the rankings again before renewing a subscription.

Source: LLM Leaderboard (Artificial Analysis) · Text to Image Leaderboard (Artificial Analysis) · Text to Video Arena (Artificial Analysis) · Text to Speech Leaderboard (Artificial Analysis) · Terminal-Bench v4.0 (Artificial Analysis) · About Suno (Suno) · Add or discover new sources for your notebook (Google Help) · Create a notebook in Gemini Notebook (Google Help)