ZMIME

Inkling 2 Mini vs Hy4 Base

Compare documented capabilities and access. A dash means a comparable figure is unavailable.

Model specifications
SpecificationInkling 2 MiniThinking MachinesHy4 BaseTencent
TaskMultimodal reasoning and codingMultimodal pretraining
Published capabilitiesReuses its middle layers three times per token. Thinking Machines reports 48.2% on DeepSWE v1.1.Pretrained checkpoint for continued training and adaptation. Tencent’s flagship Hy4 uses a 78-layer language backbone with a separate vision encoder.
Model size26.4B total / 3.4B active parameters770B total / 49B active per token in the language backbone
AccessDownloadable weights and Tinker inference and customizationDownloadable under the Tencent Hy License
Operating limitsNative mixed-precision checkpoint occupies approximately 20 GB.—
SourcesModel weights · Additional documentationModel weights
InputsText, ImagesText, Images, Video
WeightsOpen weights · Apache-2.0Open weights · Tencent Hy License
ReleasedNov 20, 2026Nov 20, 2026