Ornith 1.5 vs Qwen3.6: which agentic coding model should you use?

Ornith 1.5 is an MIT-licensed Qwen3.5 derivative in four sizes; Qwen3.6 is Alibaba's Apache 2.0 pair with image input. Compare sizes, licenses and evidence.

  • Updated

Verdict. Ornith 1.5 is a post-trained derivative of Qwen3.5 aimed at agentic coding, and Qwen3.6 is Alibaba's own newer release tuned for the same job, so test both on your repository. Ornith comes in 9B, 35B-A3B, 122B and 397B sizes under the MIT license, with FP8, NVFP4, GGUF and MLX builds. Qwen3.6 has a 35B-A3B mixture of experts and a 27B dense model under Apache 2.0, with image and video input. Every Ornith benchmark number comes from DeepReinforce, including its runs of competing models, and Qwen's numbers come from Alibaba, so neither table settles the question.

Specifications

Specifications Ornith 1.5Qwen3.6
DeveloperDeepReinforceAlibaba Qwen
ReleasedAugust 19, 2026April 16, 2026
Total parameters403B (Ornith-1.5-397B)36B (35B-A3B)
Active parameters17B3B
ArchitectureMixture of expertsMixture of experts
Context window256K tokens256K tokens
InputTextText, Image, Video
License MIT (Permissive) Apache 2.0 (Permissive)
Memory for the smallest variant (4-bit)6.66 GB (Ornith-1.5-9B)19 GB (27B)

Performance claims come from the developer and have not been independently verified.

Pick Ornith 1.5 if

  • You want a 9B model for a small GPU, or a 122B or 397B model, sizes Qwen3.6 does not offer.
  • You want the MIT license and ready FP8, NVFP4, GGUF and MLX builds.
  • Your agent harness uses Qwen-style tool-call parsers.
  • You want to try DeepReinforce's agentic post-training on the Qwen architecture.

Pick Qwen3.6 if

  • You need image or video input, which Ornith does not accept.
  • You prefer a release from the original developer under Apache 2.0.
  • You want the 27B dense model for a 24 to 48 GB GPU.

Ornith 1.5 · Qwen3.6