Alibaba
Maker · 25 entries
Everything from Alibaba
| Name | What to know | Price or memory | |
|---|---|---|---|
| strong Best open-weights image generator and editor on Arena, small enough for a home GPU | ~8 GB | ||
| mixed Strong on web-dev and agent rankings, but behind rivals on reasoning, terminal coding and vision. | $1.425 / $4.275 | ||
| promising Wan 3.0 leads Arena video editing by a narrow margin. | |||
| promising Open Qwen 3.8 runs well on modest hardware, but it is wordy and independent testing is thin | ~16 GB | ||
| promising Runs well on home hardware; one odd sandbox incident is rare and likely harmless. | ~93.7 GB | ||
| Agent mode with Qwen4-Max planned multi-step work using parallel tools (web search, Python, PDF) and produced a polished CEO-succession board briefing | |||
| Qwen 3.8 Max preview is accessible through the paid Token Plan | |||
| Naive UMAP maps of Qwen-embedded Steam reviews mostly separate by text length rather than by what the review says | |||
| mixed Open 6B image model that a newer rival reportedly beats; little direct testing. | |||
| Real-time interpreter with speaker labels and voice cloning; 60 input and 29 spoken output languages; API only | |||
| Fast omni-modal model with 1M-token context taking text, audio, image and video; API-only on Qwen platform | |||
| Open-sourced local-first search combining ripgrep, BM25 and vector search behind one CLI, for people and AI agents | |||
| Qwen3.8 Flash described as 125B parameters with 6B active and a 1M-token context window | |||
| 1.7B model run with custom kernels and rolling partial stitching to add streaming; 40-80 ms speech-to-text | |||
| Used as the higher-quality TTS with better emotional control when generation time allows | |||
| Open medical model for 2D and 3D imaging plus text: Q&A and clinical reports; 32B and 8B versions | ~72 GB | ||
| Animates a character photo from a reference video: hands, faces, multiple characters, non-human, camera angle control | ~33 GB | ||
| Pre-trained on over 38,000 hours of manipulation data, all open datasets plus human hand videos converted to robot demonstrations | |||
| Qwen-Robot suite: RobotNav navigation, RobotManip manipulation and RobotWorld video world model, each with a technical report, built on Qwen VL backbones | |||
| Video world model trained on 8.6M video-text pairs across 20+ embodiments | |||
| Alibaba cinematic video model supporting text-to-video, image-to-video and subject-to-video; reviewer says it was released around April 2026 | |||
| Production TTS from Tongyi's Qwen Audio team in two tiers, real-time Flash and higher-quality Plus, 16 languages | |||
| Generates long, character-consistent dance videos from a character image and music, with optional keyframe control | ~85 GB | ||
| v0.3 lets the live character interact with its surroundings (gestures, hair, gaze), not just portrait talking; paper only | |||
| Took #1 on Arena vision among open weights | |||
Plans
What each one gives you.Alibaba Cloud Token Plan
Alibaba Cloud Coding Plan Pro