▲Open-Weight Models Close the Gap With Closed Frontier Labs

Score history

Daily conviction score, 0 to 100. Higher means the thesis is more strongly corroborated.

Now 100

Showing the last few days. Unlock full score history.

Why the conviction moved

• Aug 28
  Strengthened +4

  Tencent's WeChat team open-sourced a 2B/4B/9B multimodal embedding family already running in production, with the 9B reported at 80.6 overall on MMEB-v3 across text, images, video and visual documents. Releasing a production-proven retrieval component with downloadable weights removes a paid closed-API dependency from the multimodal RAG stack, where embeddings are the highest-volume call.

• Aug 27
  Strengthened +6

  GLM-5.3-Flash topped OpenRouter usage while running anonymously as "Ox Alpha", meaning developers picked it over closed frontier options without knowing its lineage. Blind routing share is a harder test of parity than benchmark scores, since it reflects paid production traffic rather than leaderboard position.

Showing the last 2 days. Unlock the full record.

Source trail

• Supporting · August 28, 2026

  Tencent's WeChat Team Open-Sources a Multimodal Embedding Family Already Running in Production

  Tencent's WeChat team open-sourced a 2B/4B/9B multimodal embedding family already running in production, with the 9B reported at 80.6 overall on MMEB-v3 across text, images, video and visual documents. Releasing a production-proven retrieval component with downloadable weights removes a paid closed-API dependency from the multimodal RAG stack, where embeddings are the highest-volume call.

  AI ML Big Data (Telegram)

• Supporting · August 27, 2026

  Z.ai Releases GLM-5.3-Flash, a 320-Billion-Parameter Open-Weight Multimodal Model Under an MIT License

  GLM-5.3-Flash topped OpenRouter usage while running anonymously as "Ox Alpha", meaning developers picked it over closed frontier options without knowing its lineage. Blind routing share is a harder test of parity than benchmark scores, since it reflects paid production traffic rather than leaderboard position.

  AI ML Big Data (Telegram)