Skip to content
Source
Sign inSign up
TanStack/schemas

No fact for the maker of a hosted third-party model (GLM and DeepSeek on Mistral and BytePlus, Gemma on Gemini) #710

Send to my agent
Opentombeckenham opened this issue 0 comments
tombeckenhamcommented

Problem

Some hosts serve models made by another company. The row has no fact that says so. Snapshot 2026-11-21, chat rows.

HostRows made elsewhereExamples
mistral6glm-5-2, zai-glm-5, zai-glm-5-3, zai-glm-5-3-flash, zai-glm-latest, deepseek-v4-1-flash-1017
byteplus11deepseek-v4-pro-ga-260813, glm-5-2-260617, gpt-oss-120b-250805
gemini2gemma-4-26b-a4b-it, gemma-4-31b-it
groq6openai/gpt-oss-120b, qwen/qwen3.8-27b (the id prefix names the maker)

On these rows sameAs, openWeights, weightsUrl and providerMetadata are all null.

Impact

TanStack AI's Mistral adapter lists Mistral's own models, and its Gemini adapter leaves out Gemma. Today that needs hand-written id prefixes (glm-, zai-, gemma-). The prefixes go stale each time a host adds a family.

Ask

  • Add a fact for the model's maker when it is not the host, from the host's model page. sameAs may already be the field for this; if so, fill it.
  • Fill openWeights where the host's page states it (Gemma, gpt-oss).

Acceptance

  • Each row in the table names its maker, or has a factSources entry that says the host does not state it.
  1. tombeckenham assignedtombeckenham

  2. tombeckenham linked this to #731

  3. tombeckenham unlinked this from #731

Sign in to join this conversation.