Artificial Analysis lists Qwen3.5 122B A10B
Artificial Analysis has published a benchmark profile for Qwen3.5 122B A10B, an Apache 2.0 open-weights mixture-of-experts model that activates 10B parameters per token, placing it among mid-sized open models.
Artificial Analysis has published a release page for Qwen3.5 122B A10B. The page covers two variants, one reasoning and one non-reasoning, and compares them on intelligence, cost, speed and latency.
According to the listing, the model has 125B total parameters and 10B active parameters per token. It accepts text and image input, produces text output, and has a 262k-token context window. The weights are on Hugging Face under the Apache 2.0 license.
On latency, the non-reasoning variant had the lower time to first answer token of the two, at 4.40 seconds.
Artificial Analysis groups open-weights models by size, and at 125B total parameters this model falls in its "Medium" class (40B–150B). The model is compared only against other open models in that class. The page also lists capability indexes built from evaluations including AA-Briefcase v1.1, AA-Omniscience v1.1, HLE-Diamond, AutomationBench-AA, GDP.pdf, Terminal-Bench 4.0 and FrontierCode-AA 2.0.
The text available for this article does not include the model's actual index scores, per-token pricing, or where it ranks against other models. It also gives no release date or developer statement.