AI Models - Tier Maker
AI Models - Official Tier
Editorial placements for AI Models, with the reasoning behind each list.
AI Models - Blind Rank
The next character from AI Models is drawn for you, so you never pick who comes next. Top 5 holds one per slot. The Tier List works like the regular board, with several characters in each row.
AI Models - Versus
Run a fast elimination tournament for AI Models and keep choosing winners until only one character remains.
AI Models - Bracket
Build a championship bracket for AI Models. Drag the seeds, or fill them at random. Click a fighter to send them through, or roll the match.
AI Models - Alignment Chart
Arrange AI Models in a grid or circle with up to 16 sections. Place characters freely, including between sections.
AI Models - 0–10
Give each AI Models character a score from 0 to 10. Same scores sit together, from 10 at the top down to 0.
Number keys set 0 to 9. Tap a character on the board to change that score.
Characters (34)
The Assistant You Actually Open
A version number is a bad way to choose an assistant. ChatGPT, Claude, Gemini and Grok each get a new patch before the last one has settled, and the name on the box changes faster than the habit of opening it. This board ranks the company, not the patch. S is the short list you can hand to someone who wants one tab for writing, code and ordinary questions, from a lab that still trains its own frontier model. ChatGPT is there because most people already have the account and OpenAI still trains the line. Claude is there because Anthropic is the house people open when the paragraph or the repository has to be careful. Gemini is there because it already sits inside Search, Android and Workspace, and Google still trains it. A is the next door: a real model, plus a reason to keep the tab. Grok lives on X and will answer what the others dodge. Copilot is already in Windows, Office and GitHub. Cursor is the editor a lot of developers now leave open all day, even though most of the models inside it belong to someone else. Meta AI is already in WhatsApp, Instagram and Facebook. DeepSeek, Qwen and Kimi are the files developers keep when the frontier price is the problem. B earns one job, not every job: Mistral when the data has to stay in Europe, Perplexity when the answer needs a source attached, Z.ai for the GLM line, MiniMax as a lab you call by name, Apple Intelligence as the model inside the phone, Amazon Q for the office that already runs on AWS, and ByteDance for Doubao, which is enormous where that company's apps are the internet. Kilo Code and OpenCode are the subsidized coding plans: you open the agent, and the model is whoever is cheap this month. StepFun is the Shanghai lab behind the Step line. Tencent is Yuanbao and Hunyuan, enormous where WeChat is the internet. C is a real company with a narrower room. Cohere and AI21 sell to businesses. 01.AI is the Yi line, no longer the headline. Baidu's Ernie matters inside Baidu. NVIDIA's Nemotron matters if you already run their chips. Xiaomi's MiMo meets most people because of the phone. IBM's Granite meets people who already buy IBM. Databricks meets its own customers. Hugging Face is the shelf where the open files are stored, not a chatbot. Jev, from TypeSafe, does not chat at all: it returns a typed decision, and there is no tab to open. SenseTime, iFlytek and InternLM are real Chinese lines. Most people meet them inside those companies' own apps, not as the default tab.
Chatgpt
Claude
Gemini
Grok
Copilot
Meta Ai
Deepseek
Qwen
Kimi
Cursor
Mistral
Perplexity
Z Ai
Minimax
Apple Intelligence
Amazon Q
Bytedance
Kilo Code
Opencode
Stepfun
Tencent
Cohere
Ai21
01 Ai
Baidu
Nvidia
Xiaomi
Ibm
Databricks
Hugging Face
Jev
Sensetime
Iflytek
Internlm
Open Weights You Can Run
Some of these companies will not let the model leave the website. Others publish the weights. This board only asks about the file: can you download it, fine-tune it, or serve it without renting that company's own chat. An open-source coding app is not the same thing as an open model. DeepSeek, Qwen, Kimi and Meta AI sit in S because the public weights are the thing people take home, not a demo parked beside a closed flagship. Z.ai, the lab behind GLM, still puts a large model under a license you can read. Mistral still ships weights you can run even though its biggest branded chat stays on its own site. NVIDIA publishes Nemotron for people who already have the hardware. StepFun has published Step weights, including an Apache file and the October 2026 Step 5 Preview under the lab's own license. Tencent publishes Hunyuan weights, including Apache 2.0 files, while the Yuanbao app stays on Tencent's side. MiniMax, 01.AI, IBM, Xiaomi, InternLM, SenseTime and iFlytek have a public line. InternLM is Shanghai AI Lab's long-running open series. SenseTime has put SenseNova files out, including image models under Apache 2.0. iFlytek has put the small on-device Spark line out. None of those is the file most people mean when they say they downloaded the model. Hugging Face is the shelf those files sit on, and Databricks is the platform that wraps them. The bottom row is a closed door on purpose. ChatGPT, Claude, Gemini, Grok, Copilot, Perplexity, Apple Intelligence, Amazon Q, Cohere, AI21, ByteDance and Baidu can be excellent products. They do not hand you the file. Cursor, Kilo Code and OpenCode are clients and plans: the agent may be open, the weights are not theirs. Jev is a closed API from TypeSafe. It returns a decision. It does not give you a file.
Deepseek
Qwen
Kimi
Meta Ai
Z Ai
Mistral
Nvidia
Stepfun
Tencent
Minimax
01 Ai
Ibm
Xiaomi
Hugging Face
Databricks
Internlm
Sensetime
Iflytek
Chatgpt
Claude
Gemini
Grok
Copilot
Perplexity
Apple Intelligence
Amazon Q
Cohere
Ai21
Bytedance
Baidu
Cursor
Kilo Code
Opencode
Jev
Want to rank them yourself? Open the Tier Maker tab.
How many characters will participate?
Click a character to advance, or roll the match. An empty slot sends the other fighter through.
Available
Drag or click a character to add it. Place it over a line to share two sides. Edit or drag the labels; use arrow keys to move a focused character and Delete to return it.
