▼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 83 · -2 since Oct 11 · ranged 83 to 85

Showing the last few days. Unlock full score history.

Why the conviction moved

• Oct 10
  Strengthened +2

  Decision-1 is Alibaba's open-weight Qwen3.5-9B with Microsoft post-training. A US hyperscaler building a commercial product on a downloadable Chinese base model shows open weights doing practical production work against closed models like GPT-6 Sol.

• Oct 6
  Strengthened +3

  Nvidia-backed Reflection AI announced Beam, a 501B-parameter open-weight mixture-of-experts model (23B active) that it says scores 80.9 on SWE-bench Verified and will ship under Apache 2.0 later in October. A permissively licensed model scoring near the frontier on coding narrows the closed labs' lead on practical work, though the scores are the company's own and the weights are not yet out.

Showing the last 2 days. Unlock the full record.

Source trail

• Supporting · October 10, 2026

  Microsoft's Decision-1 scores fixed answer options in one pass and charges nothing for output tokens

  Decision-1 is Alibaba's open-weight Qwen3.5-9B with Microsoft post-training. A US hyperscaler building a commercial product on a downloadable Chinese base model shows open weights doing practical production work against closed models like GPT-6 Sol.

  AI ML Big Data

• Supporting · October 6, 2026

  Reflection AI unveils Beam, a 501-billion-parameter open-weight model built to compete with China's GLM-5.2

  Nvidia-backed Reflection AI announced Beam, a 501B-parameter open-weight mixture-of-experts model (23B active) that it says scores 80.9 on SWE-bench Verified and will ship under Apache 2.0 later in October. A permissively licensed model scoring near the frontier on coding narrows the closed labs' lead on practical work, though the scores are the company's own and the weights are not yet out.

  Reflection AI (via Hacker News)