Ornith 1.5 35B (A3B variant) is a 36B-parameter open-weight release from the ornith-ai lineage that updates and supersedes the earlier Ornith-1.0-35B model. The architecture accepts both text and image input, with multimodal support handled through a dedicated mmproj projection file that pairs with the language model weights. Community quantizations also expose speculative decoding via MTP and imatrix-aware builds, giving self-hosters concrete paths to balance latency and memory footprint on consumer GPUs.

For practitioners who prefer running weights locally, the model is readily available in GGUF form with a range of precision options, and the Q4_K_M build at roughly 22GB is highlighted as a balanced choice between size and quality. Full BF16 weights sit near 71GB for maximum fidelity, while Q8_0 and Q6_K_L variants fill the middle ground. This positions the 1.5 release as a flexible, multimodal successor aimed at users who want a capable reasoning model without relying solely on a hosted endpoint, particularly for workflows that mix text and visual context.