Logseq/Entity/AI/Model
AI Model
- In this garden, AI Model pages represent one released version of a machine learning model — Anthropic/Model/Claude/5/5/Opus, OpenAI/Model/GPT/6/Sol — named under the provider that ships it.
What counts as an instance
Naming and links
- Pages live under the provider's namespace. The provider should be an instance of Logseq/Entity/Company or Logseq/Entity/Organization. Version numbers that contain a decimal split the same way Logseq/Entity/Time/Year splits a year: most significant digit group first, then the next. Claude Opus 5.5 is Anthropic/Model/Claude/5/5/Opus, not
Claude/Opus/5.5 and not Claude/5.5/Opus. Claude Fable 5.1 is Anthropic/Model/Claude/5/1/Fable. GPT-4.1 is OpenAI/Model/GPT/4/1. - A version with no decimal keeps one number segment: OpenAI/Model/GPT/6/Sol, OpenAI/Model/GPT/6/Astra.
- The last segment is the tier or given name when the release has one — Opus, Fable, Sol, Astra, Sonnet. A name that is only a number needs no extra leaf.
- Provider namespaces already in use:
OpenAI/Model/…, Anthropic/Model/…, Google/AI/Model/…, DeepSeek/Model/…, xAI/Model/…. A new provider takes the same shape. - Older pages may still use a dotted leaf (
Claude/Fable/5.1, Claude/Sonnet/4.5) or put the tier before the number. New pages follow the split. Leave a legacy path as filed unless a rename is asked for; add alias:: on the new page only when the old path has no file of its own. - Always link the full namespace path. Bare display text — Gemini 2.5 Pro, Opus 5.5 — does not resolve and leaves the model out of the graph.
- Page names stay singular, per Logseq/Pref/Page/Name.
Frontmatter
- It should have standard entity attribution in Logseq/Entity/Frontmatter.
alias:: — the model code, the string an API call actually carries: alias:: [[claude-opus-5-5]], alias:: [[gpt-6-sol]]. Prefer the official code from the official model vendor over a prettified display name. Do not create more than one alias.tags:: — [[AI/Model]] on a page you are creating.date-created:: — the model's public release date when known, per Logseq/Date.- Shared frontmatter conventions live on Logseq/Frontmatter.
--- fences around page-level attributes are illegal in Logseq; unwrap them.
Page shape
- After the H1, embed the model's docs namespace — at minimum the Logseq/Entity/AI/Model/Doc/API page, and any Logseq/Entity/AI/Model/Doc/Analysis pages worth seeing on the model itself. That is a Logseq/Page/Embed:
{{embed [[Provider/Model/…/Doc/API]]}}{{embed [[Provider/Model/…/Doc/Artificial Analysis/Intelligence]]}}
- A thin page is an H1 plus those embeds. Anthropic/Model/Claude/5/5/Opus is the current example. Official facts live on the Doc/API page; third-party scores live on the analysis page. Do not repeat the same URL on the model H1.
- A stub, when the model is only being referenced and the official URL is not yet in hand:
- # Model Name, then - *Stub page - detailed documentation pending*. Prefer a thin page with Doc/API once the official URL is known. - A full page may still add house sections under the embeds when the model is the subject of longer notes:
## Key Features — capabilities, context window, knowledge cutoff, and the release status: stable, preview or experimental.## Performance Highlights — benchmark results with figures, linked to the benchmark's own page where one exists.## Model Tiers — the variants and how they differ, when the release has more than one.## Access Details — API availability, pricing, rate limits, usage restrictions.## Technical Specifications — architecture, training data, input and output formats, special capabilities, and the model version code.## References — papers and sources that are not the Doc/API or Doc/Analysis URLs already embedded.
Writing an instance
- Deduplicate first: model code, official docs URL, provider path, and any Logseq/Entity/AI/Leaderboard row that already names it.
- Create the model page at the split path. Create
<model>/Doc/API with an H1 link to the official docs. Create <model>/Doc/<Source>/… when a third-party card exists. Embed those pages under the model H1. - Research from those two sources, not from a separate import checklist: the official docs for price, context, reasoning, and speed; ArtificialAnalysis (or another analysis page) for independent scores.
- Give figures, not adjectives. Intelligence Index: 58 is a fact a reader can use; better at coding is not.
- Link sibling and predecessor models so a generation reads as a sequence.
- When a Logseq/Entity/AI/Leaderboard is being refreshed and this model is missing from the garden, a thin page plus Doc/API is enough. A later pass can fill house sections.
- Facts here decay. Revisit a page when benchmarks land, when pricing or availability changes, when a variant ships, or when a release moves from preview to stable.
Relationship to other types
Examples in this garden