Hugging Face tells you what exists. FixTools tells you what fits, what it costs to run, what the license can block, and how to ship it.
Stack picker
This is a directional shortlist, not a leaderboard. Your own eval set still gets the final vote.
The atlas
Curated cards contain FixTools review notes. Live Hub results stay fresh and link back to the source model card.
Qwen
A newly released multimodal mixture-of-experts model with a large context window and broad serving support.
Fast multimodal agents and coding copilots
Teams that need a small consumer-device model
Z.ai
A natively multimodal 320B MoE model with 18B active parameters and a one-million-token context window.
Long-context coding and visual agents
Consumer hardware or simple classification jobs
DeepSeek
The official V4 Flash release, tuned for agentic work and shipped with speculative decoding support.
Coding, tool use, and agentic workloads
On-device apps with modest RAM
Moonshot AI
A 2.8T open-weight native multimodal model built for coding, reasoning, and very long workflows.
Long-horizon knowledge work and visual agents
Cost-sensitive self-hosting without a cluster
IBM
A mid-size dense reasoning model with selectable thinking modes and a long context window.
Reasoning, RAG, code, and controlled enterprise stacks
Teams chasing only the largest frontier benchmark scores
Liquid AI
A compact hybrid model designed for on-device agents, with broad local-runtime support.
Private local agents on laptops and edge devices
Frontier-level open-ended knowledge work
Tencent
A universal multimodal embedding model that emits normalized 2,048-dimensional vectors.
Unified text, image, video, and document retrieval
Audio search or tiny CPU-only services
Baidu
A document-intelligence model aimed at parsing long, complex visual documents in one workflow.
Long-horizon document parsing and OCR pipelines
Zero-trust environments that prohibit custom model code
Black Forest Labs
A compact generation-and-editing image model designed for real-time use on consumer GPUs.
Fast image generation and editing in product experiences
CPU-only deployment
Krea
The distilled, few-step Krea 2 variant for commercial creative and application workflows.
Few-step creative image generation and design tools
Teams that need a tiny image model
Lightricks
An open-weight world model for local generation of synchronized high-fidelity video and audio.
Synchronized video-and-audio generation and fine-tuning
Lightweight serverless functions
MiniMax
An omni-modal generative system that understands mixed media and generates video with native stereo audio.
High-end video with synchronized stereo audio
Global launches before license review
MiniMax
A music model that generates complete stereo songs up to five minutes from lyrics and a music brief.
Long-form songs with vocals and evolving arrangements
Low-latency sound-effect generation
Index Team
A zero-shot TTS model supporting five languages, cross-lingual transfer, emotion, and speed control.
Multilingual expressive TTS and controlled voice transfer
Products without voice-consent safeguards
Qwen
A speech-recognition model with language identification across 30 languages and 22 Chinese dialects.
Multilingual transcription and Chinese dialect coverage
Tiny always-on edge microphones
Why this wins
FixTools should not mirror millions of repositories or pretend every checkpoint is production-ready. The moat is continuously maintained decision context.
Memory class, deployment target, and local-vs-cluster reality.
Clear, custom, gated, and territory-limited terms made visible.
Pinned commands, runtimes, quantizations, and known compatibility traps.
Use-it-for, skip-it-for, comparisons, and eval-ready shortlists.
Use FixTools for PDFs, JSON, images, metadata, and the developer utilities around your AI stack.
