Entity Template

Unified field structure for the five entity collections — Models, Companies, Agents, APIs, Benchmarks (plus the pricing time-series, which is a per-provider price snapshot rather than an entity). This document is the single source of truth for how a record in data.sinoaihub.com is shaped, and it is what the backfill scripts enforce.

Every entity record carries a common core of identity + provenance fields, then a collection-specific set. Unknown or non-applicable values use null / unknown / not publicly disclosed — never invented.

Common core (every entity)

Field	Meaning	Notes
`id`	Stable lowercase-hyphen identifier	Equals the filename stem; immutable
`name`	Display name	H1 title on the page
`type`	`model` \| `company` \| `agent` \| `api` \| `benchmark` \| `pricing`	New in 2B-P1; added to all 59 records
`provider`	Owning organization (→ companies id)	Models / agents / APIs / pricing link to a company; companies link to themselves implicitly
`status`	`active` \| `preview` \| `deprecated` \| `discontinued`	Models carry it explicitly; other collections infer `active` from `last_verified` freshness
`release_date`	ISO 8601 (or `~YYYY-MM`)	Models have it; companies use `founded`; agents / APIs / benchmarks usually lack a public date
`official_url`	Canonical main-site page	The "Canonical page" link at the top of every record
`sources`	Evidence list	Unified schema — see evidence-layer.md
`last_verified`	Verification date	Present on every record
`verification_status`	`verified` \| `partially_verified`	New in 2B-P1; derived from source confidence + documented disclosure gaps

Collection-specific fields

Models

Field	Example	Required
`model_family`	DeepSeek-V4	yes
`version`	`0813`	when documented
`aliases`	`deepseek-v4-pro-0813`	when documented
`architecture`	MoE: 1.6T total / 49B active	when disclosed
`parameter_information`	`total_parameters`, `active_parameters`	when disclosed
`context_window`	1048576	when disclosed
`maximum_output`	393216	when disclosed
`capabilities`	reasoning / coding / math / vision / tool_calling / function_calling / structured_output	yes (each tri-state)
`open_weight`	true / false	yes
`license`	MIT \| Apache-2.0 \| proprietary \| custom	yes
`self_hosting`	true / false	yes
`api_available`	true / false	yes
`pricing`	input/output per 1M + currency + pricing_ref	yes (if api_available)
`official_api`	true / false	yes
`cloud_providers`	`[DeepSeek Platform]`	when documented
`regions`	`[china, global]` \| `unknown`	when documented
`benchmark_results`	benchmark / score / metric / date / source	when documented
`release_history`	date / event / source / verification_date	when multi-event
`model_versions`	version / date / source	when documented (2B-P1)
`license_changes`	change events	none documented (2B-P1)
`known_limitations`	list	yes

Companies

Field	Example	Notes
`aliases`	深度求索
`founded`	2023	replaces `release_date`
`headquarters`	Hangzhou, Zhejiang, China
`funding`	"No external funding disclosed…"
`ai_products`	list
`foundation_models`	→ models ids
`open_models`	→ models ids
`agents`	→ agents ids
`api`	→ apis ids
`major_releases`	name / date / type
`timeline`	date / event / type / source / verification_date	company milestones (2B-P1)
`open_source_projects`	list
`official_documentation`	URL
`official_website`	URL

Agents

Field	Example	Notes
`company`	→ companies id
`description`	1–3 factual sentences
`agent_type`	`coding` \| `browser` \| `computer_use` \| …	defensible
`underlying_models`	→ models ids
`framework`	TypeScript terminal agent
`tool_calling` / `browser_use` / `computer_use` / `mcp` / `memory` / `planning` / `multi_agent` / `api`	true / false	tri-state
`pricing`	description
`deployment`	`cloud` \| `self_hosted` \| `both`
`open_source`	true / false
`license`	MIT
`github` / `documentation`	URL
`use_cases` / `limitations`	list

APIs

Field	Example	Notes
`provider`	→ companies id
`api_type`	`official` \| `cloud_hosted` \| …
`endpoint`	https://api.deepseek.com
`authentication`	Bearer API key
`streaming` / `function_calling` / `tool_calling` / `structured_output` / `vision`	true / false
`context_limits`	model / input_limit / output_limit
`rate_limits`	text
`regions` / `cloud_providers`
`pricing_ref`	→ pricing id
`documentation`	URL
`api_changes`	date / change / source	DeepSeek only (2B-P1)

Benchmarks

Field	Example	Notes
`description`	what it measures
`task_type` / `dataset_size` / `evaluation_method` / `scoring`	methodology
`evaluations`	benchmark / model / score / model_version / metric / date / source_type / source_url	per-result source_type mandatory
`relevant_models`	→ models ids + scores
`limitations`
`benchmark_changes`	before/after values	none sourced (2B-P1)

Field presence matrix (2B-P1)

Counts of the 59 entity records (21 models + 6 companies + 10 agents + 6 APIs + 6 pricing + 10 benchmarks) after backfill.

Collection	Records	Common-core `type` added	`verification_status` added	Sparse (missing ≥1 optional field)
Models	21	21	21	13 `partially_verified` (context / architecture / parameters / capabilities / benchmarks not disclosed)
Companies	6	6	6	0
Agents	10	10	10	0
APIs	6	6	6	1 `partially_verified` (minimax structured_output)
Pricing	6	6	6	0
Benchmarks	10	10	10	0

Backfill rules

Values are copied only from the main-site frontmatter or existing data-hub content. Nothing is sourced from memory.
A field absent from both is left out or marked unknown / not publicly disclosed — it is never guessed.
verification_status = verified when core facts rest on high-confidence official sources and no core field is documented as undisclosed; partially_verified when a core field is documented as not publicly disclosed (the gap is stated, not hidden).