0xMarioNawfal(@RoundtableSpace):Ornith-1.5 just dropped - a family of open-source models spanning 9B Dense, 35B MoE, and 397B MoE, trained with end-to-end self-improvement. Here's what Ornith-1.5 actually looks like: > 86.1 on Terminal-Bench 2.1 > 86 on SWE-Bench verified, 79.6 multilingual > 81.4 on ClawEval, 71.2 on Tool Decathlon - Performance comparable to Claude Opus 4.8 across reasoning, agentic, and coding tasks - The model proposes its own tasks, generates scaffolds, and produces solution rollouts for RL. - FP8, GGUF, MLX, and NVFP4 quantized versions all released under MIT Open source. Commercial use unrestricted. Self-improving out of the box.

2026.08.27 18:15

Ornith-1.5 just dropped - a family of open-source models spanning 9B Dense, 35B MoE, and 397B MoE, trained with end-to-end self-improvement. Here's what Ornith-1.5 actually looks like: > 86.1 on Terminal-Bench 2.1 > 86 on SWE-Bench verified, 79.6 multilingual > 81.4 on ClawEval, 71.2 on Tool Decathlon - Performance comparable to Claude Opus 4.8 across reasoning, agentic, and coding tasks - The model proposes its own tasks, generates scaffolds, and produces solution rollouts for RL. - FP8, GGUF, MLX, and NVFP4 quantized versions all released under MIT Open source. Commercial use unrestricted. Self-improving out of the box.