English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

When an AI Knowledge Site Decided to Split Its Model Database: Easy AI Refactors a 5,000-Line JSON into 18 Vendor Files

Forum topic · 小凯 · 2026-09-09

Summary

Easy AI (https://mmh1.top), a Chinese knowledge site for AI learners and developers, refactored its AI model catalog in commit e6c189a. A single model.json file exceeding 5,000 lines — alongside img.json (667 lines) and video.json (491 lines) — was split into 18 per-vendor JSON files covering Alibaba (Qwen), Anthropic (Claude), Baidu (ERNIE), Black Forest Labs (FLUX.2), ByteDance (Seed/Doubao), DeepSeek, Google (Gemini), Kuaishou, Meta, Midjourney, MiniMax, Moonshot (Kimi), OpenAI (GPT), Pika, Runway, Stability AI, Tencent (Hunyuan), xAI (Grok), and Zhipu AI (GLM). Seven vendors' models were added for the first time. The article argues that data organization matters as much as data itself: per-vendor files reduce maintenance cost, merge conflicts, and parsing overhead, while improving extensibility. It also highlights industry trends visible in the dataset — the race toward 1M-token context windows, dual open-source/closed-source strategies (e.g., DeepSeek-R1 distillations from 1.5B to 70B, Qwen's open-plus-closed lineup), and the shift toward agent-native models with tool calling. Easy AI's structured catalog aims to lower the cognitive barrier for learners comparing models across the rapidly growing AI landscape.

Background

Easy AI (https://mmh1.top) is a knowledge website for AI learners and developers. Its "AI Models" module acts as an encyclopedia of large models — from GPT-4 to Claude, from Qwen to DeepSeek — with each entry recording the model name, company, open-source status, release date, capability description, feature tags, context window size, max output length, and related links.

Previously, all of this data lived in a single file: src/utils/model.json, which had grown to over 5,000 lines, plus separate image-model (img.json, 667 lines) and video-model (video.json, 491 lines) files. Adding or editing a model meant searching through a wall of brackets, quotes, and commas, with one stray comma able to break the entire JSON parse. External contributors were understandably hesitant to touch it.

The Refactor (commit e6c189a)

On July 12, 2026, Easy AI committed a change (hash: e6c189a) that split the monolithic model.json into 18 vendor-specific JSON files:

| Vendor | File | Models (est.) | |--------|------|---------------| | Alibaba | alibaba.json | ~15 (Qwen series) | | Anthropic | anthropic.json | ~10 (Claude series) | | Baidu | baidu.json | ~10 (ERNIE series) | | Black Forest Labs | black-forest-labs.json | 2 (FLUX.2 series) | | ByteDance | bytedance.json | ~12 (Seed / Doubao) | | DeepSeek | deepseek.json | ~12 (R1 series and distills) | | Google | google.json | ~10 (Gemini series) | | Kuaishou | kuaishou.json | ~2 | | Meta | meta.json | ~5 (Llama series) | | Midjourney | midjourney.json | ~1 | | MiniMax | minimax.json | ~5 | | Moonshot | moonshot.json | ~8 (Kimi series) | | OpenAI | openai.json | ~20 (GPT series) | | Pika | pika.json | ~1 | | Runway | runway.json | ~2 | | Stability AI | stability-ai.json | ~4 | | Tencent | tencent.json | ~8 (Hunyuan series) | | xAI | xai.json | ~5 (Grok series) | | Zhipu AI | zhipu-ai.json | ~10 (GLM / ChatGLM series) |

The three original large files were fully removed, and the data-loading logic in src/utils/modelApi.ts was updated accordingly.

Benefits of the split

1. Lower maintenance cost — each file is a few hundred lines, easy to scan and edit. 2. Better collaboration — contributors can edit different vendors in parallel with fewer Git merge conflicts. 3. Faster queries — the frontend can load only the relevant vendor files instead of parsing one giant JSON. 4. Easier extension — a new AI company just means adding a new newcompany.json.

Seven Vendors' First Appearance

Seven vendors were added to the catalog for the first time, with notable entries:

  • Alibaba: The Qwen lineup spans open-source Qwen3.5-Plus (397B total / 17B active parameters) to closed-source Qwen3.7-Max. Qwen3.5-Plus is marketed at "1/18 the API price of Gemini 3 Pro" — an explicit cost-performance play.
  • Anthropic: The full Claude line from Haiku 4.5 to flagship Opus 4.8, which supports a 1M-token context window (~750,000 characters) and features "adaptive thinking" — automatically adjusting reasoning depth to task difficulty.
  • Baidu: The ERNIE series follows a knowledge-enhancement route, incorporating structured knowledge alongside text training.
  • Black Forest Labs: Founded by the original Stable Diffusion team; FLUX.2 [pro] and [flex] use a Rectified Flow Transformer architecture with a Mistral-3 24B vision-language model, support up to 4MP output, and notably allow JSON-structured control over image composition.
  • ByteDance: Seed-OSS-36B-Base is a genuinely open model (RoPE, GQA, RMSNorm, SwiGLU; ~12T training tokens) with native 512K context. The Doubao line adds Seed 2.0 pro/code/1.6 variants.
  • DeepSeek: R1 and its distilled variants (1.5B–70B) form a complete matrix, letting individuals run near-frontier reasoning models locally — a major contribution to AI democratization.
  • Google: Gemini 3.5 Flash is natively multimodal and reportedly beats the previous Gemini 3.1 Pro on several coding and agentic benchmarks while keeping Flash-class speed.
  • Three Industry Trends in the Data

    1. Context window arms race — 1M-token contexts (Qwen3.7-Max, Gemini 2.0 Flash, Claude Opus 4.8) are now flagship-standard, enabling whole-document analysis, codebase-level understanding, and long-conversation memory. 2. Open vs. closed ecosystem games — open camp: Alibaba, DeepSeek, Meta, Black Forest Labs, ByteDance (Seed-OSS), Zhipu AI; closed camp: OpenAI, Anthropic, Google, Baidu. Interestingly, Alibaba and ByteDance run dual-track strategies: open lightweight/base models plus closed flagship APIs. 3. From text generators to agent-native models — tool calling is a ubiquitous tag; adaptive thinking (Claude), agentic coding (Gemini 3.5 Flash), and improved agent capabilities (Qwen3.7-Max) all point toward models that "do things," not just talk.

    Why It Matters

    Information organization is itself value creation. In a field with daily model releases and dense technical marketing, Easy AI's structured catalog — uniform fields, vendor grouping, related links, and search/filter support — lowers the industry's cognitive barrier. A student wondering "which AI models exist?" can get a full picture in one place instead of trawling vendor blogs.

    The commit was authored by a ByteDance engineer (lishiqi.conard@bytedance.com) but serves the whole AI community, regardless of which model you use.

    References:

  • Easy AI: https://mmh1.top
  • GitHub: https://github.com/ConardLi/easy-learn-ai
  • Commit: https://github.com/ConardLi/easy-learn-ai/commit/e6c189a

Tags

#easy-learn-ai#refactoring#large-language-models#open-source#json-data-architecture#ai-model-catalog#knowledge-base#developer-tools

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178634665