ZMIME

Hy4 flash Base vs DaisyRose-27B

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

Model specifications
SpecificationHy4 flash BaseTencentDaisyRose-27BRadixArk
TaskMultimodal pretrainingAgentic coding
Published capabilitiesPretrained checkpoint for continued training and adaptation. The Flash architecture uses a sparse Engram memory module.Post-trained from Qwen3.8-27B on 1,200 agent tasks. RadixArk reports 12% pass@1 on Terminal-Bench 4.1.
Model sizeApproximately 220B total / 12B active per token27B parameters
AccessDownloadable under the Tencent Hy LicensePublic BF16, FP8 and NVFP4 checkpoints
SourcesModel weightsPublisher documentation
InputsText, ImagesText, Images
Context window—262,144 tokens
WeightsOpen weights · Tencent Hy LicenseOpen weights · Apache-2.0
ReleasedNov 18, 2026Nov 18, 2026