Ornith-1.5 just dropped and it's actually insane 🔥
Open-source models hitting Claude Opus 4.8 levels across reasoning, coding, and agentic tasks:
9B Dense / 35B MoE / 397B MoE
Benchmarks:
86.1 Terminal-Bench 2.1
86 SWE-Bench verified (79.6 multilingual)
81.4 ClawEval / 71.2 Tool Decathlon
The alpha? End-to-end self-improvement - model literally proposes its own tasks, generates scaffolds, and produces solution rollouts for RL. No hand-holding.
FP8, GGUF, MLX, NVFP4 quantized versions all live. MIT license. Unrestricted commercial use.
This is what decentralized AI actually looks like. No walled gardens, no API rate limits, no corporate guardrails.
If you're building agents or coding tools, this changes the economics overnight.
Open-source models hitting Claude Opus 4.8 levels across reasoning, coding, and agentic tasks:
9B Dense / 35B MoE / 397B MoE
Benchmarks:
86.1 Terminal-Bench 2.1
86 SWE-Bench verified (79.6 multilingual)
81.4 ClawEval / 71.2 Tool Decathlon
The alpha? End-to-end self-improvement - model literally proposes its own tasks, generates scaffolds, and produces solution rollouts for RL. No hand-holding.
FP8, GGUF, MLX, NVFP4 quantized versions all live. MIT license. Unrestricted commercial use.
This is what decentralized AI actually looks like. No walled gardens, no API rate limits, no corporate guardrails.
If you're building agents or coding tools, this changes the economics overnight.
