Question 1

What is a complete list of AI models in 2026?

Accepted Answer

This catalog tracks every meaningful AI model across five categories: large language models (GPT-5, Claude, Gemini, Llama, Qwen, DeepSeek, GLM, Grok, Mistral, Kimi…), image generation models (Flux, Imagen, GPT-Image, Recraft), video generation (Veo, Sora, Runway, Kling), text-to-speech (ElevenLabs, OpenAI TTS) and speech-to-text (Whisper, Deepgram). Use the category nav above to filter, or the search box to jump to a specific AI model.

Question 2

What are the best AI models right now?

Accepted Answer

The "best" AI model depends on the task. For frontier reasoning, GPT-5, Claude Opus, Gemini 3 Pro and Grok 4 lead public benchmarks. For coding agents, sort by the Coding Index. For open-weight self-hosting, Llama, Qwen and DeepSeek are the strongest. The leaders summary above the catalog names the current winner per axis. Full rankings live on the LLM Leaderboard.

Question 3

What is the cheapest AI model that's still good?

Accepted Answer

Combine the Price filter with the Reasoning Index filter. The cheapest models with frontier-grade reasoning are typically open-weight (Qwen, DeepSeek, GLM) hosted by inference providers like Together, Fireworks or Groq. For closed models, the GPT-5 nano / mini, Claude Haiku and Gemini Flash families are the cost leaders.

Question 4

Which AI models support images, audio, or video?

Accepted Answer

Use the Modality filter. For image input, GPT-5, Claude (vision), Gemini and Llama 4 are the strongest. For audio input, GPT-5 with audio, Gemini and Whisper handle it natively. For video generation, Veo, Sora, Runway and Kling are the leading models; for video analysis, Gemini and GPT-5 vision can process frames.

Question 5

What are the different types of AI models?

Accepted Answer

The catalog groups AI models into five practical types: language models for text, code, reasoning and tool use; image generation models for text-to-image and image editing; video generation models for text-to-video and image-to-video; text-to-speech models for voice synthesis; and speech-to-text models for transcription. Within LLMs there are further sub-types — reasoning models (o-series, DeepSeek-R1, Claude with thinking), multimodal models (GPT-5, Gemini, Claude vision) and open-weight models (Llama, Qwen, DeepSeek, Mistral, GLM).

Question 6

What's the difference between an AI model and an LLM?

Accepted Answer

AI model is the umbrella term for any machine-learning model. LLM (large language model) is a specific kind of AI model trained on huge text corpora to predict the next token — that's what powers chatbots, code assistants and most agent products today. All LLMs are AI models, but not all AI models are LLMs: image generators (Flux, GPT-Image), video models (Veo, Sora), speech models (Whisper, ElevenLabs) and embedding models are AI models that aren't LLMs.

Question 7

What is an AI model?

Accepted Answer

An AI model is a program — almost always a deep neural network — trained on data to perform a task: generating text, images, video or speech, classifying inputs, or transcribing audio. Modern AI models are dominated by large language models (LLMs) like GPT-5, Claude, Gemini and Llama, which generate text from a prompt. This catalog tracks every meaningful AI model across text, image, video, speech and transcription with live pricing and benchmark scores.

Question 8

Where can I find open source AI models?

Accepted Answer

Filter the catalog by Openness → Open to see every model with publicly released weights — Llama, Qwen, DeepSeek, Mistral, GLM, Gemma and more. Open-source AI models can be self-hosted, fine-tuned and inspected, so they're the right fit for sensitive workloads. The Open LLM Leaderboard ranks them head-to-head.

Question 9

How is AI model pricing calculated?

Accepted Answer

For LLMs, Price is a blended per-1M-token estimate (8 parts input, 1 part output) — that mirrors a chat and agent workload where prompts, tool traces and retrieved documents dominate output. Use this catalog as an LLM price comparison across OpenAI, Anthropic, Google, Meta, xAI, DeepSeek, Qwen and other providers. Other modalities use their native unit: per image, per second of video, per minute of audio, per 1M characters for TTS. All prices come from public provider price lists and are cross-checked against billing samples through the LLM Stats proxy.

Question 10

How do I compare LLMs side by side?

Accepted Answer

Use the LLM comparison tool to compare two models by benchmarks, context window, API pricing, latency, throughput, release date, license and providers. For broader rankings, use the LLM Leaderboard.

Question 11

What's the difference between the Reasoning and Coding indexes?

Accepted Answer

Both are composite TrueSkill ratings that aggregate public benchmark results into one comparable score per model. The Reasoning Index focuses on math, science, logic and multi-step problem solving — GPQA Diamond, AIME, MATH. The Coding Index focuses on software engineering — SWE-Bench Verified, LiveCodeBench, HumanEval. Higher is better. See the LLM Stats Score methodology for the full computation.

Question 12

How are latency and throughput measured?

Accepted Answer

Latency is the median time-to-first-token over recent traffic flowing through the LLM Stats proxy; throughput is the median tokens-per-second once streaming begins. Both are rolling averages over the last 7 days, so a model that just launched will show fewer samples until it warms up.

Question 13

How often are scores and prices updated?

Accepted Answer

Pricing and provider availability refresh hourly. Latency and throughput refresh every few minutes from live proxy traffic. Reasoning and Coding indexes recompute nightly as new benchmark results come in.

Where the data comes from

Reasoning & knowledge

Coding

Multimodal

Methodology