ainewsblitz.com

Breaking

Ornith-1.5 open-weight family ships 9B, 35B MoE and 397B MoE with self-improvement training

  • Foundation Models
  • Open Source
  • Software Dev & Coding

Ornith, the DeepReinforce team behind Ornith-1.0, has released Ornith-1.5, an MIT-licensed open-weight family spanning a 9B dense model, a 35B mixture-of-experts (MoE) model that activates about 3B parameters per token, and a 397B MoE flagship trained with an end-to-end self-improvement loop. On the team’s own evaluations, Ornith-1.5-397B scores 86.1 on Terminal-Bench 2.1 (Terminus-2) and 86 on SWE-bench Verified, figures it presents as comparable to Claude Opus 4.8 on several coding and agentic tests.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.

$20
Read this article
$29/month
Unlimited — all 27,494 articles, the full archive, and comprehension quizzes
Save 72%
$98/year
≈ $8.17/month
Unlimited, billed once a year