Open-source Ornith-1.5 release: 397B MoE beats Claude Opus 4.8 on four benchmarks
rohanpaul_ai · x · 2026-08-20
Ornith-1.5 is an open-source LLM family spanning 9B Dense, 35B MoE and 397B MoE, trained with self-improving strategies. The 397B model beats Claude Opus 4.8 on four benchmarks:
- Terminal-Bench 2.1: 86.1 vs 85
- SWE-bench Verified: 86 vs 85.8 (Pro 65.1, Multilingual 79.6)
- WideSearch: 80.8 vs 72.9
- BrowseComp: 86.6 vs 84.3
- Also: DeepSWE 56, HLE 44.6, ClawEval 81.4, Tool Decathlon 71.2
The architectural jump: task generation, agent design and solution generation live in the same RL loop — the model proposes its own tasks, writes the grading scaffold, produces solution rollouts, and propagates reward back through all three stages via GRPO. This self-generated curriculum searches near the model's capability frontier, removing much human task design.
Related event: Ornith-1.5 Release: 397B MoE Self-Improvement Targets Claude Opus(24 posts)→
More from Models
- Meta open-sources Muse Spark Open agent model with day-0 TPU support in SGLang — BanghuaZ · 2026-10-16
- Saying "load-bearing" in your prompt makes Opus 5.5 stop holding back — zacharynado · 2026-10-16
- Americans get their Claude accounts banned while on business trips in China — dotey · 2026-10-16
- I priced every Grok Bot call: $20 to $200 plan in 5 days, long jobs eat 52% — jonathanmalkin · 2026-10-16
- NVIDIA releases Nemotron-3.5-Embed-1B, a multimodal embedding model covering 34 languages — jacek2023 · 2026-10-16
- EpochAI pegs GPT-6 training at up to 1.8e27 FLOP; scaling01 calls param estimates way too high — scaling01 · 2026-10-16