Model Card and System Card Index by Provider: Where Each Provider Publishes Its Cards

Every large language model provider publishes documentation about its models, but each one puts it somewhere different, calls it something different and dates it differently, so finding the right card for the model you use takes longer than reading it. This index records, for 21 providers, where model cards and system cards are published, what form they take, and which 2026 language model cards each provider's own index lists, with a link to every source. It is for developers, procurement and AI governance teams who need the official document behind a model before approving it. For what a card is and how to read one, start with Model Cards and System Cards Explained; this page is the lookup table that sits beside it. Providers are listed alphabetically; order is never a ranking. Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider.

Key takeaways

• There is no common home for cards. Of the 21 providers checked, 14 publish their main card for open-weight models as a Hugging Face README, while others use their own documentation sites, cloud user guides, PDFs or a dedicated safety hub.
• "Model card" and "system card" are used loosely. Anthropic titles its page "Model system cards", OpenAI publishes system cards and addenda on its Deployment Safety Hub, and xAI's Grok 4.20 document is titled "System Card" while its Grok 4.5 and 4.6 documents are titled "Model Card".
• Three providers keep a dated index of every card. Anthropic (23 system cards back to Claude 2), Google DeepMind (55 model cards across Gemini, Gemma, generative media and robotics) and OpenAI (Deployment Safety Hub updates) each list documents with dates on one page.
• Technical specifications often sit elsewhere. OpenAI, Meta and xAI put context windows and ids on developer model pages, and Mistral AI and Amazon Bedrock put them on per-model documentation cards, so a safety document and a specification page may both be needed.
• Addenda change the record without a new card. OpenAI publishes addenda (for example GPT-6.1 Sol and GPT-6.1 Terra as addenda to the GPT-6 Astra system card) and xAI's Grok 4.5 and 4.6 cards carry revision dates, so record the document date and revision you relied on.

Status key

Dates are the dates shown on each provider's own index or document. Anthropic's index gives months only; Google's gives an "Updated" date; OpenAI's gives a publication date; xAI's Grok 4.5 and 4.6 cards carry a date and a revision date.

Where each provider publishes cards

As of 2026-11-15. One row per provider, identical fields. "Where cards live" names the location; "Notes from the page checked" records only what that page shows.

Upstage also publishes Hugging Face cards (for example Solar Open 2 250B). For licence terms of the open-weight checkpoints above, see the Open-Weight Model Licence Tracker.

2026 language model cards by provider

As of 2026-11-15. Scope, applied the same way to every provider: documents dated 2026 that cover a language model, as listed on the provider's own dated index or published as a dated document. Image, audio, video and robotics cards are excluded for every provider. Newest first.

Anthropic

Google

OpenAI

The hub's update list ends with a "View more" link; older 2026 entries sit behind it.

xAI

Providers whose cards live per checkpoint on Hugging Face or per model on documentation pages (the other 17 providers in the first table) do not publish a single dated index of cards on the pages checked, so their 2026 cards are reached through each model's repository or documentation page linked above and through the provider hubs, such as DeepSeek Models, Mistral AI Models and Alibaba Qwen Models.

What to watch next

• Cards for the newest releases: Anthropic lists a system card for Claude Fable 5.2 and Mythos 5.2 (November 2026) and Google lists Gemini 4 Carbon and Hydrogen cards (10 and 12 November 2026); check for revisions in the weeks after release.
• xAI Grok 4.7: the docs.x.ai models page links a Grok 4.7 model page and announcement; a Grok 4.7 card was not on the pages checked.
• OpenAI addenda: GPT-6.1 Sol and Terra are documented as addenda to the GPT-6 Astra system card, so a new model may arrive as an addendum rather than a stand-alone card.
• Revision dates: xAI's Grok 4.5 and 4.6 cards carry revision dates within a week of the original; Google's index shows "Updated" dates, which can move after first publication.
• EU training content summaries: providers of general-purpose AI models in the EU publish a separate public summary of training content; see the EU General-Purpose AI (GPAI) Code of Practice Guide.

How to use this tracker

1. Find the provider row first. The first table tells you whether to look on Hugging Face, a documentation site, a cloud user guide or a safety hub.
2. Match the document to the exact model id. Cards often cover two configurations (for example Fable and Mythos) or a family; addenda cover later versions. Confirm the id you run is named in the document.
3. Record date and revision. Store the card URL, its date and any revision date in your model register next to the API id and licence.
4. Pair the card with the specification page. Where the card is a safety document, take context window, pricing and lifecycle from the developer page or provider hub; see the LLM Release Tracker and LLM Deprecation and Retirement Tracker.
5. Feed it into approval. Attach the card to your change record; the AI Model Change Approval Template and Model Card and System Card Template show what to capture.

Practitioner checklist

• For every model in production, store the provider's card or system card URL, document date and revision date.
• Check that the document names the exact model id and configuration you use, including addenda for later versions.
• For open-weight checkpoints, save a copy of the Hugging Face card and licence file at the repository commit you deploy.
• Where a provider shows "Not stated" for a system card, record which other documents (technical report, model page) you relied on instead.
• Re-check the provider's index when a model is updated under the same name or alias.
• Map the card's contents to your AI bill of materials; see AI BOM vs Model Card vs Datasheet.
• Confirm each row against the provider source before relying on it.

Standards and references

• Anthropic: Model system cards (accessed 2026-11-15)
• Google DeepMind: Model cards (accessed 2026-11-15)
• OpenAI: Deployment Safety Hub (accessed 2026-11-15)
• xAI: Grok 4.6 Model Card (accessed 2026-11-15)
• Amazon Web Services: Amazon Bedrock models at a glance (accessed 2026-11-15)
• Mistral AI: Models overview (accessed 2026-11-15)
• Microsoft AI: MAI Thinking 1 (accessed 2026-11-15)
• DeepSeek: DeepSeek-V4.1-Flash-1017 model card (accessed 2026-11-15)

Related NHI Mgmt Group resources: Model Cards and System Cards Explained · Model Card and System Card Template · Open-Weight Model Licence Tracker · Frontier AI Models Compared · AI BOM vs Model Card vs Datasheet