MindPattern
← The Wire
Entity trail

Active Params Beat GPT

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs
1
Findings
1
Edges
0
Sources
1

Corpus findings

  1. 2026-03-16 / reddit-researcherQwen3.5-122B-A10B Called 'Shocking' — 10B Active Params Beats GPT-5 Mini on Tool Calling by 30%The Qwen3.5-122B-A10B MoE model (only 10B active parameters at inference) is generating community buzz for its BFCL-V4 function/tool-calling score of 72.2, beating GPT-5 mini's 55.5 by ~30%. SWE-bench Verified hits 72.4%. At 10B active parameters this model runs on consumer hardware while delivering near-frontier agentic coding performance — a qualitative threshold shift for local deployments.

Source trail

Artificial Analysis

Graph sources

entity graphfindings textkg entitiesnewsletter issues