According to CoinMeta, the Ornith-1.5 series has added a new 122B MOE model, which is based on the Qianwen 3.5 architecture and leads in some programming self-tests compared to the Qwen3.8-Flash-Next. This series was first released in August of this year and included three other specifications: 397B, 35B, and 9B. This time, an intermediate specification has been added. The new model has its weights made available, offering versions such as FP8, NVFP4, GGUF, and MLX. The Ornith-1.5-122B continues to use the model architecture of Qwen3.5-122B-A10B, with a total of about 122B parameters. Each inference activation uses approximately 10B parameters and supports 256K native contexts, image understanding, tool calls, and inference. Official self-tests show that the 122B model scored 84.5 out of 100 in SWE-Bench and verified (as well as non-hack), which is higher than the 82.8 points achieved by the Qwen3.8-Flash-Next model.