The Open Public Registry for Frontier AI Models & Research Checkpoints
ModelRegistry is a community-driven, machine-readable index tracking the state of foundational artificial intelligence. Rather than letting outdated models clutter developer workflows or burying flagship LLMs under niche audio drops, ModelRegistry maintains a clear dual-tier structure:
- Primary Foundation Flagships: The reigning general-purpose models developers actually use in production (e.g. Meta Muse Spark 1.3, ChatGPT 5.6, Anthropic Claude Fable 5.1, Google Gemini 3.8 Flash, DeepSeek V4.1 Flash).
- Latest Specialized Checkpoints: Newly dropped breakthroughs (e.g. Gemini 3.8 Live Extended Thinking, ChatGPT Images 2.5, Meta Muse Voice Transcribe, Hy4 Preview).
Hosted at modelregistry.tirup.in.
| Laboratory | Primary Flagship | Latest Checkpoint | Context | Access | Pricing |
|---|---|---|---|---|---|
| Anthropic | Claude Opus 5.5 | Claude Sonnet 5.5 | 1M | Proprietary | $4 in / $20 out |
| OpenAI | GPT-6.1 Sol | GPT-6 Luna | 1M | Proprietary | $2 in / $10 out |
| Google DeepMind | Gemini 4 Argon | Gemini 3.8 Live Extended Thinking | 1M | Proprietary | $2 in / $10 out |
| xAI | Grok 4.7 | Grok Voice Transcribe 2.0 | 500k | Proprietary | $2 in / $6 out |
| DeepSeek | DeepSeek V4.1 Flash | DeepSeek V4-Pro (0813) | 1M | Open (MIT) | $0.3 in / $1.2 out (Open) |
| Meta AI | Muse Spark 1.3 | Muse Voice Transcribe | 262k | Open (Meta Community) | $0.05 in / $0.15 out (Open) |
| Alibaba Cloud (Qwen) | Qwen3.8 2.4T A95B | Qwen3.8 Flash | 1M | Open (Qwen Community) | $0.8 in / $2.4 out (Open) |
| Mistral AI | Mistral Large 4 | - | 1M | Open (Open Weights (27 Oct 2026)) | $1.36 in / $4.18 out (Open) |
| Tencent Hunyuan | Hy3 | Hy4 Preview | 262k | Open (Tencent Hunyuan Community) | $0.13 in / $0.53 out (Open) |
| Z.ai | GLM 5.3 | GLM 5.3 Flash | 1M | Proprietary | $0.9 in / $2.7 out |
| MiniMax | MiniMax M3 | - | 1M | Proprietary | $0.3 in / $1.2 out |
| NVIDIA | Nemotron 3 Ultra 550B | - | 262k | Open (NVIDIA Open Model) | $0.6 in / $2.4 out (Open) |
| Xiaomi MiMo | MiMo-V2.6 Pro | MiMo-V2.6 Flash | 1M | Open (MIT) | $0.435 in / $0.87 out (Open) |
| Moonshot AI | Kimi K3 | - | 262k | Proprietary | $3 in / $15 out |
| Kuaishou Kling | Kling 3.0 | - | - | Proprietary | $0.084 per second |
| Runway | Runway Gen-4.5 | - | - | Proprietary | $0.15 per second |
| Sarvam AI | Sarvam 105B | - | 128k | Open (Apache 2.0) | $0.33 in / $0.83 out (Open) |
| TypeSafe AI | Jev | - | - | Proprietary | $0.042 in / $0 out |
- Dual-Tier Model Organization: Immediate distinction between heavyweight general foundation models and newly trained checkpoints.
- SOTA Domain Leaderboard: Head-to-head verified evaluations across Reasoning, Agentic Coding, Context Capacity, and Inference Value.
- Open Telemetry & Syndication:
GET /api/v1/models- Public JSON REST API with filtering parameters.GET /api/v1/models/{id}- Single-record fetch for polling one model.GET /api/v1/changes?since=YYYY-MM-DD- Incremental changelog sync.- Full human reference with examples: modelregistry.tirup.in/docs.
GET /rss.xml- Live RSS 2.0 syndication feed for newly registered models.GET /llms.txt- Machine-readable ground truth formatted for AI answer engines and web crawlers.
Query the registry directly from your terminal without installing anything:
curl -s https://modelregistry.tirup.inEmbed real-time frontier flagship badges directly in your GitHub READMEs:
[](https://modelregistry.tirup.in)
[](https://modelregistry.tirup.in)
[](https://modelregistry.tirup.in)
[](https://modelregistry.tirup.in)
[](https://modelregistry.tirup.in)ModelRegistry provides free, unauthenticated REST endpoints for bots, CLI tools, and agent workflows:
# Fetch all registered models
curl -s https://modelregistry.tirup.in/api/v1/models
# Fetch only primary company flagships
curl -s "https://modelregistry.tirup.in/api/v1/models?flagshipOnly=true"
# Fetch only open-weight community models
curl -s "https://modelregistry.tirup.in/api/v1/models?openWeights=true"
# Filter by laboratory
curl -s "https://modelregistry.tirup.in/api/v1/models?company=anthropic"
# Fetch one model without pulling the full registry
curl -s https://modelregistry.tirup.in/api/v1/models/gpt-6-astra
# Incremental sync: changelog entries since a date, then fetch changed records
curl -s "https://modelregistry.tirup.in/api/v1/changes?since=2026-10-01"Full parameter reference, live counts, and Python/JS examples: modelregistry.tirup.in/docs.
ModelRegistry uses a single-file contribution workflow. You only ever edit one file: data/models.ts.
- Add your model to
data/models.ts. - Run
pnpm test- it validates the schema and auto-syncs this README table. - Open a Pull Request!
See CONTRIBUTING.md for the copy-paste snippet.
Paste this into any AI agent (Claude, Cursor, Codex, Copilot) and it will walk you through contributing:
You are helping me contribute a new AI model to ModelRegistry (https://modelregistry.tirup.in, repo: https://github.com/TirupMehta/ModelRegistry), the open registry of frontier AI models.
Follow this workflow step by step. Ask me for any fact you cannot verify from an official source - never invent specifications, benchmarks, pricing, or dates.
1. Clone and set up:
git clone https://github.com/TirupMehta/ModelRegistry.git
cd ModelRegistry
pnpm install
2. Open data/models.ts and append ONE object to the modelsData array using this exact schema (every field required unless marked ?):
{
id: "lab-model-name-0102", // unique lowercase kebab-case id
companyId: "openai", // one of: anthropic, openai, google, xai, deepseek, meta, qwen, mistral, tencent, z-ai, minimax, nvidia, xiaomi, moonshotai, kuaishou, runway, sarvam, typesafe
companyName: "OpenAI", // lab display name
name: "Model Display Name",
version: "1.0", // lab version string
releaseDate: "2026-09-08", // YYYY-MM-DD, first public availability
isCompanyFlagship: false, // true ONLY if this is the lab's primary flagship (exactly 1 per lab - demote the previous flagship to false)
isLatestCheckpoint: true, // true if this is the lab's newest release
statusBadge: "NEW DROP", // short uppercase pill, e.g. "NEW DROP", "OPEN WEIGHTS", "EXPIRES SEPT 10"
category: "flagship", // one of: reasoning | flagship | audio | open-weights | multimodal | code | image | video
categoryLabel: "Human Readable Label",
contextWindow: "1,048,576 tokens", // human string; visual models use descriptive windows like "8s clips"
contextWindowTokens: 1048576, // sortable number; use 0 for non-token windows
maxOutputTokens: "65,536 tokens",
parameters: "1.6T MoE", // architecture; write "Undisclosed (...)" when the lab published nothing - never fabricate
openWeights: false,
license: "Proprietary API", // e.g. "MIT License" for open weights
pricing: { input: 10.0, output: 50.0 }, // USD per 1M tokens; per-second video models add pricingUnit: "per second"
highlight: "One factual sentence: release date plus what changed.",
modalities: ["Text", "Vision"], // any of: Text | Vision | Code | Audio | Video | Image
benchmarks: {}, // lab-published scores only (sweBench, mmluPro, gpqa); {} when none published
links: {
announcement: "https://...", // MANDATORY: official announcement, docs page, paper, or verified weights repo
playground: "https://...", // ? optional chat/API playground
apiDocs: "https://...", // ? optional API docs
weights: "https://..." // ? optional weights repo
}
}
3. If the model is from a laboratory not yet tracked, also add it to data/companies.ts with: id, name, shortName, description, website, headquarters, accentColor, latestFlagship.
4. Freshness sweep (STRICT - never skip): the registry must never contradict itself.
a. Exactly ONE isCompanyFlagship:true per lab - demote the previous flagship to false.
b. Scrub stale superlatives on the entries this release dethrones (same lab first, plus any cross-lab record it takes): #1, NEWEST, SOTA, best, latest, reigning, most advanced, newly. Rewrite those badges/highlights in past-neutral terms.
c. A record belongs ONLY to its verified current holder - never copy a crown onto the newcomer without an official source.
d. Update data/companies.ts latestFlagship / latestReasoning / description when they changed.
e. Update data/leaderboard.ts spotlights if the newcomer takes a spotlight slot.
f. Keep every highlight to 1-2 tight lines; trim any older highlight that grew into a paragraph.
5. Run: pnpm test
This validates the schema and auto-syncs the README table. Fix every error it reports.
6. Commit on a new branch and walk me through opening the Pull Request (use gh if authenticated).
RULES:
- Official source required for every fact. No rumors, leaks, or benchmark guesses.
- Popularity bar: top-15 OpenRouter weekly volume, a primary flagship, or a genuinely frontier capability. No obscure checkpoints or minor variants.
- Freshness is mandatory: a submission that adds a model without demoting/scrubbing what it replaced will be rejected.
- Touch ONLY data/models.ts (plus data/companies.ts for a new lab). Website, API, RSS, and README update automatically.
# Clone the repository
git clone https://github.com/TirupMehta/ModelRegistry.git
cd ModelRegistry
# Install dependencies
pnpm install
# Start development server
pnpm run devOpen http://localhost:3000 in your browser.
MIT © Tirup Mehta & ModelRegistry Contributors. See LICENSE for details.