When AI Models Grew From a Phonebook Into a Library: 243 Models, 19 Companies, One Restructured Database

Forum topic · 小凯 · 2026-08-19

Summary

A recent commit to the open-source easy-learn-ai project restructured its AI model registry, splitting a single monolithic model.json into 19 per-company files covering 243 models from 2023 to 2026. The article uses this refactor as a lens to map the explosive growth of the AI industry. It categorizes the ecosystem into three battlefields: text LLMs (Qwen, Claude, GPT, Gemini, DeepSeek, Kimi, GLM), image generation (FLUX, Midjourney, Stable Diffusion, SeedEdit), and video generation (Pika, Runway, Kling, Doubao). It highlights China’s nine-player ecosystem—Alibaba, Baidu, ByteDance, DeepSeek, Moonshot, Zhipu, MiniMax, Tencent, Kuaishou—now contributing over 100 models. The piece argues that organized, structured data turns reference material into a decision-making tool for developers choosing APIs based on region, task, budget, context length, and privacy.

Key Points

Scale of the ecosystem: The easy-learn-ai repository now tracks 243 AI models across 19 companies, split from a single monolithic file into one file per vendor (Alibaba, Anthropic, Baidu, Black Forest Labs, ByteDance, DeepSeek, Google, Kuaishou, Meta, Midjourney, MiniMax, Moonshot, OpenAI, Pika, Runway, Stability AI, Tencent, xAI, Zhipu).
The metaphor: The industry has evolved from a small-town phonebook (one GPT-4 in early 2023) into a multi-floor library requiring structured navigation.
Three competitive battlefields:
Text LLMs — Qwen, Claude, GPT, Gemini, DeepSeek, Kimi, GLM competing on reasoning, context length, code, and agentic capabilities.
Image generation — FLUX, Midjourney, Stable Diffusion, SeedEdit forming a mature industrial pipeline since 2022.
Video generation — Pika, Runway, Kling, Doubao teaching AI to understand time across 24-frame coherent sequences.
Model leaderboard (by count in the repo): OpenAI 38, Alibaba 28, ByteDance 20, DeepSeek 17, with Black Forest Labs punching above its weight at just 2 models (FLUX dominating image generation).
China's "group army": 9 Chinese companies contribute over 100 models collectively. DeepSeek-R1 proved small teams can build frontier models with open-source + low-cost strategy; Qwen iterated to 3.7; Kimi pioneered 2M-token context; ByteDance tightly couples models with Doujin/CapCut.
Why the refactor matters: Splitting by company enables horizontal comparison—the ability to study each vendor's strategic choices (breadth vs. depth, open vs. closed, text-only vs. multimodal).
Decision framework for developers: Model selection now depends on region (latency/compliance), task type, budget, context length, and data privacy tolerance.
Technical detail: The change touched src/utils/modelApi.ts (55 lines modified), the API layer bridging frontend to backend data structures.

---

Why This Matters

The restructuring reflects a deeper shift: AI is no longer a single "thing" but an ecosystem—comparable to how electricity evolved from light bulbs to motors, heaters, and networks. When knowledge organization evolves, the way you use that knowledge must evolve too. The library is built and shelved; readers now decide which sections to explore.

---

*Commit: e6c189a | easy-learn-ai model database refactor and expansion* *Models tracked: 243 | Companies covered: 19 | Data span: 2023–2026*

When AI Models Grew From a Phonebook Into a Library: 243 Models, 19 Companies, One Restructured Database

Summary

Key Points

Why This Matters

Tags