SYLIN

Tencent

Maker · 18 entries

Everything from Tencent

NameWhat to knowPrice or memoryRanks
HunyuanImage InstructModelmixed Third-best open image editor on Arena, but well behind the top two.#3 image editing (open weights)
HunyuanVideoModelmixed Third among open video models on Arena, far behind the leader.#3 video generation (open weights)
Hy FlashModelpromising Very cheap and good enough for everyday work, but not frontier-class$0.035 / $0.14
HyModelmixed Strong and cheap on security tasks, but one user saw broken responses.~229 GB
DSpineModelpromising Speeds up Qwen3.8-27B by roughly 4-5x and edges out DFlash 2.~78 GB
Hy-SimulMTModelTencent 30B MoE (3B active) simultaneous Chinese-English translation with voice cloning; GGUF Q4 ~20 GB, full 61 GB; Apache 2.0~61 GB
Hy-NomasModelunproven Open real-time world model from Tencent; quality claims are untested by others~34 GB
WeVisOne-VLModelunproven Open vision model for detection, OCR and segmentation; only Tencent's benchmarks so far~19 GB
MulticsModelMultiplayer world model giving each player a synced camera view; trained on 4 players, scales to 10; hierarchical communication cuts compute 86%~16 GB
GAEModelGeometry-native autoencoder generates video with depth, camera path and 3D reconstruction for consistent worlds; weights under 12 GB~12 GB
WorldCrafterModelTencent's interactive world model with implicit 3D point-cloud memory; code and weights released~147 GB
AuKModelTencent-affiliated open speech generation and editing model: TTS, cloning, word edits, enhancement, emotion and timbre change~6.1 GB
EVIEModel4.5B visual document retriever that matches page images to text queries with late interaction, no OCR step~10 GB
ScopeFrameworkCamera-path control for image-to-video, built on Wan 2.2 and DiffSynth Studio; code released
WorldClawModelMulti-agent system generating open 3D worlds from text, coarse to fine; no open-source commitment stated
Hunyuan3D BuffaloModelUnified 3D model that generates, text-edits and segments into parts; code and models marked coming soon
AngelSlimFrameworkTencent's open-source compression toolkit: quantization, speculative sampling, sparse attention, token compression, distillation; also image, video and 3D models
HunyuanImageModelTook #1 on Arena text-to-image among open weights