OpenBMB
Vintage: 2026-09. First dated OpenBMB page in this thread. Evidence is 2026-09-07-openbmb-minicpm5-2b-open-release-aa-index-15 (
method: grok-bot;x_video: false). Official @OpenBMB X + fetched Hugging Face / Artificial Analysis pages. ModelScope / openbmb.cn linked, not body-fetched.
OpenBMB
One-line summary: Issuer lab behind minicpm5-2b (7 Sep 2026 Apache-2.0 dense ~2.5–2.6B on-device LM). Launch post claimed Artificial Analysis Intelligence Index 23; later the same day OpenBMB thanked AA and restated Index v4.2 = 15.
What it is
A Chinese open-weight lab (X: @OpenBMB; weights on Hugging Face). This page is the lab home for the MiniCPM5-2B drop. It is not a rewrite of moonshot-ai / thinking-machines-lab / liquid-ai.
Why it matters to this thread
Open-weight issuer labs and on-device releases are in-scope. MiniCPM5-2B is the first citable OpenBMB event in this wiki. The load-bearing measurement dispute (Index 15 vs launch 23) lives on minicpm5-2b.
Key facts (from 2026-09-07-openbmb-minicpm5-2b-open-release-aa-index-15)
- From 2026-09-07-openbmb-minicpm5-2b-open-release-aa-index-15 (@OpenBMB, 2026-09-07): announced MiniCPM5-2B as open source; launch post claimed Intelligence Index 23 and Agentic Index 20.
- From the same source (@OpenBMB): later the same day quote-thanked @ArtificialAnlys and restated Index v4.2 = 15.
- From the same source (@OpenBMB): opened training-data / recipe / RL-stack pieces (UltraData-Code ~550B tokens; UltraData-SFT-Agent-2609 ~500K samples; UltraData-RL-2609 80K+ samples; UltraX ~100B tokens / 114M samples; Meshy; JustRL II). Named artifacts, not new product pages.
- From the same source: thanked SGLang and vLLM for day-0 support. No new serving-engine pages.
What this source does not establish
- No lab valuation, founder bio, or prior MiniCPM lineage in this pass.
- ModelScope / openbmb.cn not body-fetched.
- Did not re-file gemini-3-8-flash.
- No video.
x_video: false.
Contradictions / tensions
- Launch Index 23 vs later thank-you 15 (AA-aligned). Prefer 15 for Index grain; keep 23 on the record. Full treatment: minicpm5-2b.
Open questions
- How this lab sits relative to moonshot-ai / Z.ai on frontier vs on-device open weights — not answered here.