THE DECODER (Jul 16, 2026) coverage of Inkling release.

Benchmarks (Artificial Analysis):

  • Artificial Analysis Intelligence Index: 41 — leading U.S. open-weights model
  • Above Nemotron 3 Ultra (38), Gemma 4 31B (29), gpt-oss-120b (24)
  • Still trails leading Chinese open models and proprietary Claude/GPT on overall quality
  • Strong agentic performance cited
  • Weak factual accuracy: AA Omniscience +2; accuracy ~40%; hallucination rate ~63%
  • Multimodal: text, images, audio

Architecture recap: 975B MoE / 41B active; competitive with DeepSeek V4 / GLM 5.2 / Kimi K2.6 class size but not quality leadership.