Ornith-1.5 Open-Source Model Family Released, 397B MoE Benchmarks Against Claude Opus 4.8

On August 19, the open-source Ornith-1.5 model family was officially released in three configurations—9B Dense, 35B MoE, and 397B MoE—under the MIT license, offering a range of scales for different inference needs and resource constraints.

Confirmed

• The models were first released by @tarruda; the three variants (9B Dense, 35B MoE, 397B MoE) and the MIT license are official information.
• The core technical highlight is an end-to-end self-improvement training strategy: as relayed by @rohanpaulai, the model extends the self-scaffolding framework, autonomously proposing new tasks, generating task-specific scaffolds, and producing reinforcement learning solutions—continuously creating its own learning experiences rather than just passively completing tasks.
• Official evaluations claim SOTA among open-source models of the same size on reasoning, Agent, and coding tasks, with performance approaching or even matching Claude Opus 4.8; @rohanpaulai noted that the 397B MoE version has specific benchmark results on Terminal-Bench and others.

Unconfirmed

• Claims like "on par with Claude Opus 4.8" currently come mainly from the publisher's own evaluations and promotion, with no independent third-party replication yet; reposts from @iamfakhrealam and @aftahiai merely repeat the same talking points, adding no new independent evidence.

Why It Matters

• End-to-end self-improvement (the model generating its own learning tasks and scaffolds) is a new paradigm distinct from conventional post-training; if replicated, it could reduce dependence on manually constructed training data.
• The MIT license plus full coverage from 9B up to 397B MoE makes it accessible to everyone from individual developers to large-scale deployments—a direct challenge from the open-source camp to top closed-source models.

2026-08-19 ~ 2026-08-20 · 10 related posts

• [source] Ornith 1.5 released: 9B dense and 35B/397B MoE models — tarruda · 2026-08-19
• Open-Source Ornith-1.5 Matches Claude Opus in Reasoning and Coding — aftahi_ai · 2026-08-19
• Open-source Ornith-1.5 released under MIT, claims Claude Opus-level performance — iamfakhrealam · 2026-08-19
• Open-source Ornith-1.5 hits SOTA with self-improving strategies — aftahi_ai · 2026-08-19
• [source] Ornith-1.5: Self-Improving Models Match Claude Opus Performance — rohanpaul_ai · 2026-08-19
• Ornith-1.5 open-source family launches, 397B MoE claims Claude Opus 4.8 parity — KokaOP · 2026-08-19

4 near-duplicate retellings: aftahi_ai · aftahi_ai · testingcatalog · Shruti_0810