Qubrid Models Registry

A unified, provider-organized registry of frontier AI models - built for gateways, agents, and inference infrastructure.

One registry. Every major provider. Zero guesswork.

Explore Models · Categories · Contributing · Roadmap

---

Table of Contents

• Why This Exists
• Features
• Quick Stats
• Supported Providers
• Model Categories
• Featured Models
• Repository Structure
• Example Folder Structure
• Use Cases
• Contributing
• Roadmap
• Built By Qubrid

---

Why This Exists

The AI model landscape is fragmented. Model IDs, capabilities, and release cadences differ across OpenAI, Anthropic, Google, Qwen, DeepSeek, and dozens of other labs - each with its own naming scheme, documentation, and API surface.

Qubrid Models Registry solves that fragmentation with a single, predictable layout:

Problem	How this registry helps
Scattered provider docs	One tree: providers/{Provider}/{Family}/{model}
Inconsistent discovery	Browse by vendor, product line, or capability category
Hard-to-scale integrations	Stable paths for gateways, SDKs, and automation
No shared mental model	Same hierarchy for every provider

Whether you are routing traffic in production, benchmarking frontier models, or scaffolding an agent framework - start here.

---

Features

Provider-first layout	Every model lives under its vendor - no flat, unmaintainable lists
Standardized depth	Provider → Family → Model - three levels, every time
Multi-provider coverage	13 labs, 62 frontier models, one repository
Discoverability	Logical folders map to how teams actually talk about models
Scalable architecture	Add providers, families, or models without restructuring the tree
Infrastructure-ready	Designed for sync jobs, metadata layers, and API gateways

---

Quick Stats

Metric	Count
Total models	62
Providers	13
General LLMs	35
Coding models	5
Vision-language models	5
Thinking / reasoning models	4
OCR models	1
Open-weight models	10

---

Supported Providers

Provider	Families	Models	Focus
OpenAI	GPT-5, GPT-4, GPT-OSS	7	Frontier + OSS
Anthropic	Opus, Sonnet, Haiku	6	Claude 4.x lineup
Google	Gemini	4	Pro & Flash tiers
Qwen	General, Coder, VL, Thinking, Open-Weight	23	Broadest coverage
DeepSeek	V3, V4, Reasoning	6	Efficiency + reasoning
Moonshot AI	Kimi	4	Long-context agents
NVIDIA	Nemotron	3	Enterprise Nemotron
Mistral AI	Mistral	1	Open-weight instruct
Meta	Llama	1	Llama 3.3
Tencent	Hunyuan	1	OCR
MiniMax AI	MiniMax	1	M2.5
Microsoft	Fara	1	Agentic 7B
ZAI	GLM	2	GLM-4.7 & GLM-5

---

Model Categories

Models are grouped by capability, not just vendor. Use these categories when routing, benchmarking, or tagging workloads.

General LLMs - chat, completion, and broad instruction-following

Default frontier models for production assistants, RAG pipelines, and general-purpose agents.

Examples: gpt-5.4, claude-opus-4-7, gemini-2.5-pro, Qwen3-Max, DeepSeek-V3.2, Kimi-K2.6

Typical paths: providers/OpenAI/GPT-5/, providers/Anthropic/Opus/, providers/Qwen/General/

Coding Models - software engineering, patches, and repo-scale context

Optimized for code generation, refactoring, and IDE/agent integrations.

Examples: Qwen3-Coder-Plus, Qwen3-Coder-Next, Qwen3-Coder-480B-A35B-Instruct

Path: providers/Qwen/Coder/

Vision-Language Models - image + text understanding and multimodal agents

Unified perception and language for document AI, UI agents, and visual QA.

Examples: Qwen3-VL-Plus, Qwen3-VL-235B-A22B-Instruct, gemini-3.1-pro-preview

Path: providers/Qwen/Vision-Language/

Thinking / Reasoning Models - extended deliberation and chain-of-thought

Higher-latency, higher-depth inference for math, planning, and complex tool use.

Examples: DeepSeek-R1-0528, Kimi-K2-Thinking, Qwen3-Next-80B-A3B-Thinking

Paths: providers/DeepSeek/Reasoning/, providers/Qwen/Thinking/, providers/MoonshotAI/Kimi/

OCR Models - document text extraction and layout understanding

Specialized vision pipelines for scanned documents and structured capture.

Example: HunyuanOCR → providers/Tencent/Hunyuan/HunyuanOCR/

Open-Weight Models - self-hosted and on-prem inference

Run on your GPUs, your VPC, or your sovereign cloud - no proprietary lock-in.

Examples: Qwen3.5-397B-A17B, gpt-oss-120b, Llama-3.3-70B-Instruct, Mistral-7B-Instruct-v0.3

Paths: providers/Qwen/Open-Weight/, providers/OpenAI/GPT-OSS/, providers/Meta/Llama/

---

Featured Models

Model	Provider	Path	Best for
GPT-5.4	OpenAI	providers/OpenAI/GPT-5/gpt-5.4	Flagship production workloads
Claude Opus 4.7	Anthropic	providers/Anthropic/Opus/claude-opus-4-7	Maximum capability Claude
Gemini 2.5 Pro	Google	providers/Google/Gemini/gemini-2.5-pro	Multimodal Google stack
Qwen3-Max	Qwen	providers/Qwen/General/Qwen3-Max	Top-tier general Qwen
DeepSeek-V4-Pro	DeepSeek	providers/DeepSeek/V4/DeepSeek-V4-Pro	Next-gen efficiency
Kimi-K2.6	Moonshot AI	providers/MoonshotAI/Kimi/Kimi-K2.6	Long-context agents

---

Repository Structure

Every model is a leaf directory under a consistent three-level hierarchy:

providers/
└── {Provider}/
    └── {Family}/
        └── {model-id}/

Full registry tree (click to expand)

providers/
├── Anthropic/
│   ├── Opus/          → claude-opus-4-7, claude-opus-4-6, claude-opus-4-5
│   ├── Sonnet/        → claude-sonnet-4-6, claude-sonnet-4-5
│   └── Haiku/         → claude-haiku-4-5-20251001
├── DeepSeek/
│   ├── Reasoning/     → deepseek-r1-distill-llama-70b, DeepSeek-R1-0528
│   ├── V3/            → DeepSeek-V3, DeepSeek-V3.2
│   └── V4/            → DeepSeek-V4-Flash, DeepSeek-V4-Pro
├── Google/
│   └── Gemini/        → gemini-3.1-pro-preview, gemini-3-flash-preview, …
├── Meta/
│   └── Llama/         → Llama-3.3-70B-Instruct
├── Microsoft/
│   └── Fara/          → Fara-7B
├── MiniMaxAI/
│   └── MiniMax/       → MiniMax-M2.5
├── MistralAI/
│   └── Mistral/       → Mistral-7B-Instruct-v0.3
├── MoonshotAI/
│   └── Kimi/          → Kimi-K2.6, Kimi-K2-Thinking, …
├── NVIDIA/
│   └── Nemotron/      → NVIDIA-Nemotron-3-Super-120B-A12B, …
├── OpenAI/
│   ├── GPT-OSS/       → gpt-oss-120b
│   ├── GPT-4/         → gpt-4o, gpt-4o-mini, gpt-4.1
│   └── GPT-5/         → gpt-5.4, gpt-5.4-mini, gpt-5.4-nano
├── Qwen/
│   ├── Coder/
│   ├── General/
│   ├── Vision-Language/
│   ├── Thinking/
│   └── Open-Weight/
├── Tencent/
│   └── Hunyuan/       → HunyuanOCR
└── ZAI/
    └── GLM/           → GLM-4.7, GLM-5

---

Example Folder Structure

Add a new model - create a leaf folder under the correct provider and family:

providers/OpenAI/GPT-5/gpt-5.4/
providers/Anthropic/Opus/claude-opus-4-7/
providers/Qwen/Coder/Qwen3-Coder-Plus/
providers/DeepSeek/V4/DeepSeek-V4-Pro/

Navigate programmatically:

provider  = "Qwen"
family    = "Vision-Language"
model_id  = "Qwen3-VL-Plus"
path      = f"providers/{provider}/{family}/{model_id}"

Gateway routing pseudo-config:

routes:
  - id: frontier-general
    path: providers/OpenAI/GPT-5/gpt-5.4
  - id: frontier-reasoning
    path: providers/DeepSeek/Reasoning/DeepSeek-R1-0528
  - id: coding
    path: providers/Qwen/Coder/Qwen3-Coder-Plus

---

Use Cases

Use case	How teams use this registry
AI gateways	Map stable paths → upstream provider endpoints
Routing systems	Policy-based model selection by category and SLA
Benchmarking	Fixed catalog for eval harnesses and leaderboards
Agent frameworks	Discover reasoning vs. general models per task
Inference platforms	Seed deployment manifests and model cards
SDK generation	Auto-generate typed enums from folder structure
Playgrounds	Populate model pickers with verified IDs

---

Contributing

We welcome contributions - new providers, families, model folders, and documentation improvements.

1. Fork the repository
2. Create a branch - add/provider-model-name
3. Add the model under providers/{Provider}/{Family}/{model-id}/
4. Open a pull request with provider, model ID, and category

Adding a provider

providers/NewProvider/
└── ProductLine/
    └── model-id/

Keep names aligned with official provider documentation. Prefer lowercase kebab-case for API-style IDs (e.g. claude-opus-4-7) and preserve vendor casing for branded releases (e.g. DeepSeek-V4-Pro) where already established in this registry.

Contribution checklist

• Model folder is a leaf (no nested model dirs)
• Path follows providers/Provider/Family/model
• README stats updated if model count changes
• Category documented in PR description

---

Roadmap

Phase	Planned capability
Metadata	model.yaml per leaf - version, modality, release date
Pricing	Reference token pricing and batch discounts
Benchmarks	Linked eval scores (MMLU, HumanEval, etc.)
Context windows	Max input/output tokens per model
API compatibility	OpenAI-compatible, Anthropic, Gemini flags
Multimodal tags	Text, image, audio, video capability matrix
Automated syncing	CI pipeline to validate against provider APIs

---

Built By Qubrid

Qubrid builds AI infrastructure for teams shipping production inference - unified access to frontier and open-weight models, with the operational depth startups and enterprises expect.

This registry is the canonical model catalog behind Qubrid's gateway, routing, and deployment stack. Open-sourced so the community can discover, integrate, and extend the same foundation.

---

If this registry saves you time, star the repo - it helps others find it.

62 models · 13 providers · 1 tree

Maintained by Qubrid AI · MIT License