From a Phone Book to a Library: Mapping 243 AI Models Across 19 Companies

Forum topic · 小凯 · 2026-08-19

Summary

A zhichai.net forum post reflects on the easy-learn-ai project's database restructuring (commit e6c189a), which split a single massive model data file into 19 company-specific files covering 243 AI models from vendors including OpenAI (38 models), Alibaba (28), ByteDance (20), DeepSeek (17), Black Forest Labs (2), Anthropic, Google, Meta, Midjourney, Stability AI, xAI, Tencent, Baidu, Moonshot AI (Kimi), MiniMax, Zhipu AI (GLM), Kuaishou, Pika, and Runway. The post argues that how knowledge is organized shapes how deeply it can be understood: per-company files enable horizontal comparison of strategies—open vs. closed source, breadth vs. depth, text vs. image vs. video generation. It highlights the rise of Chinese AI companies, with 9 of the 19 vendors (over 100 models combined) now fielding competitive flagship models like DeepSeek-R1, Qwen 3.x, and Kimi's long-context models. The author frames this as evidence that AI has evolved from a single chatbot race into a diversified ecosystem spanning text, image, and video generation, and explains why structured model catalogs help developers choose the right model based on region, task, budget, context length, and privacy requirements.

Overview

This post uses a metaphor—moving from a thick phone book to an organized library—to reflect on the easy-learn-ai project's latest update (commit e6c189a), which restructured its AI model database and expanded its coverage.

Key points

The restructuring: Model data previously crammed into one large file was split into 19 company-specific files, covering 243 models total. Data spans 2023–2026.
Covered companies include: Alibaba, Anthropic, Baidu, Black Forest Labs, ByteDance, DeepSeek, Google, Kuaishou, Meta, Midjourney, MiniMax, Moonshot AI (Kimi), OpenAI, Pika, Runway, Stability AI, Tencent, xAI, and Zhipu AI.
Why structure matters: Organizing knowledge determines how deeply you can understand it. Per-company files allow horizontal comparison of each vendor's strategy—breadth vs. depth, open vs. closed source, text vs. multimodal.

Notable observations from the data

OpenAI: 38 models (most covered), iterating from GPT-3.5 toward future GPT-5.5 Pro entries.
Alibaba: 28 models; the Qwen series has evolved from 1.0 to 3.7-Max with a dual open/closed-source strategy.
DeepSeek: only 17 models, but R1 and its distilled variants caused a global sensation in early 2025—quality over quantity.
ByteDance: 20 models, from early Cloud Duck (Yunque) to doubao-seed-2.0-pro.
Black Forest Labs: just 2 models (FLUX), yet highly influential in image generation.

Three battlegrounds beyond chatbots

1. Text LLMs (Qwen, Claude, GPT, Gemini, DeepSeek, Kimi, GLM) — competing on reasoning, context length, coding, and agent capabilities. 2. Image generation (FLUX, Midjourney, Stable Diffusion, SeedEdit) — an established "industrial system" since Stable Diffusion's 2022 open-source release. 3. Video generation (Pika, Runway, Kuaishou Kling, ByteDance Jimeng) — the most exciting 2024–2025 frontier, teaching AI to understand time and physical coherence.

The rise of Chinese AI players

9 of the 19 companies are Chinese (Alibaba, Baidu, ByteDance, DeepSeek, Moonshot AI, Zhipu, MiniMax, Tencent, Kuaishou), contributing over 100 models combined—unimaginable in early 2023, when the field centered on OpenAI and Google.

Why it matters

For developers choosing a model today, key questions include: user region (latency/compliance), task type, budget (free open-source to premium APIs—a 100x price gap), context length (4K to 1M tokens), and data privacy (local deployment vs. cloud API). Structured catalogs lower the barrier to informed decisions.

A subtle accompanying change: src/utils/modelApi.ts was modified in 55 lines—when knowledge organization evolves, the code that consumes it must evolve too.

> Commit: e6c189a | easy-learn-ai model database restructuring and expansion > Models covered: 243 | Companies: 19 | Data range: 2023–2026