Ornith-1.5 claims Claude Opus 4.8 performance
alexcovo_eth · x · 2026-08-20
The Ornith-1.5 family (9B Dense, 35B MoE, 397B MoE) was released, claiming state-of-the-art performance among open-source models and parity with Claude Opus 4.8 in reasoning and agentic tasks via end-to-end self-improvement. A demo shows the 9B model running a full agentic loop offline to analyze revenue from spreadsheets.
More from coding & agent
- PRL framework adds adaptive concurrency to boost RL training efficiency — samsja19 · 2026-08-20
- OpenOutreach: open-source self-hosted AI agent that finds leads and sends emails — tom_doerr · 2026-08-20
- Grok Build Updates: Workflows Catalog, Permission Controls, and Reliability Fixes — elonmusk · 2026-08-20
- Enable a 1M token context window in Codex with gpt-5.6-sol — dfinke · 2026-08-20
- System design exercise: map components to code, execution, and storage locations — DuaneJRich · 2026-08-20
- Improv engine runs, agents now cut and assemble video riffs — nptacek · 2026-08-20