Ornith-1.5 Open Models Released, Claiming Claude Opus Performance

alejandroll10 · x · 2026-08-26

The Ornith-1.5 family of open-source LLMs has been released, featuring 9B Dense, 35B MoE, and 397B MoE variants trained with end-to-end self-improving strategies. It claims performance comparable to Claude Opus 4.8 across reasoning, agentic, and coding tasks, achieving scores like SWE-Bench (Verified 86) and DeepSWE (56). Weights are now available in BF16, GGUF, NVFP4, and FP8 formats, with MLX versions updated for Apple Silicon.

Original post →

More from Models

Models channel →