Ornith-1.5 Is a Free Open-Source Model That Claims to Rival Claude Opus 4.8 - Startup Fortune
Ornith-1.5 Is a Free Open-Source Model That Claims to Rival Claude Opus 4.8 - Startup Fortune
摘要: Home
Ornith-1.5 Is a Free Open-Source Model That Claims to Rival Claude Opus 4.8
Ornith-1.5 Is a Free Open-Source Model That Claims to Rival
正文
Ornith-1.5 Is a Free Open-Source Model That Claims to Rival Claude Opus 4.8
Ornith-1.5 Is a Free Open-Source Model That Claims to Rival Claude Opus 4.8
Janet Harrison
Aug 20, 2026 · 10:13 PM
4 min read
997 reads
Ornith-1.5 is current because Ornith announced it on X on August 20, 2026, but the safest version of this story is narrower than the published draft. The release is real, the benchmark claims are vendor-reported, and the outside-attributed comparison to Claude Opus 4.8 needed to go.
Ornith has turned a small open-model release into a much bigger claim: three new coding models, a self-improvement loop, and benchmark scores that the lab says put its largest system near Claude Opus 4.8. You should read that carefully. The models are open, the license is permissive, and the numbers are striking, but the numbers still come from Ornith's own announcement.
According to Ornith's X post on August 20, the Ornith-1.5 family includes a 9B dense model, a 35B mixture-of-experts model, and a 397B MoE flagship. The lab says all three are trained with self-improving strategies and released under the MIT license, including quantized versions in FP8, GGUF, MLX and NVFP4 formats. That part is the practical hook. If the model files are usable at the sizes Ornith is promising, developers don't have to wait for a closed API just to test an agentic coding system on their own stack.
That's the real story.
The numbers need a skeptical read
Ornith says the 397B version scored 86.1 on Terminal-Bench 2.1, 86 on SWE-Bench Verified, 65.1 on SWE-Bench Pro and 79.6 on SWE-Bench Multilingual. It also reported 56 on DeepSWE, 44.6 on Humanity's Last Exam, 81.4 on ClawEval and 71.2 on Tool Decathlon. Those are not small claims. They put Ornith-1.5 in the same conversation as the strongest closed coding agents, at least on the benchmarks the lab chose to publish.
Microsoft's MAI-Image-2.5-Pro Just Topped an AI Image Editing Leaderboard
Microsoft's MAI-Image-2.5-Pr