The Open Source AI Ecosystem
Open source is having a moment. Chinese labs (DeepSeek, Qwen, Z.ai/Zhipu) alongside Meta, Mistral and others keep pushing frontier models out as open weights — and catching up fast on the benchmarks. When a June 2026 US directive abruptly forced Anthropic to suspend its Fable 5 model, open weights jumped to the top of the agenda for corporates and governments weighing control, continuity and sovereignty."Open" in the AI era is contested, so we grade it: the model layer is split by how open each release really is, from permissively-licensed open weights to source-available and restricted licences. We also flag a few proprietary-but-independent coding tools (Cursor, Conductor, Windsurf…) — not open source themselves, but key enablers that put open models in developers' hands. Read this as an ecosystem map of the AI-era open stack, not an OSI-only landscape.
Source-available & restricted models
Research & specialty open-weight models
Inference, serving & gateways
Agent frameworks & platforms
Persistent & autonomous agents
Coding assistants & devtools
Search & relevance engines
Data pipelines & transformation
Labeling, feedback & data governance
LLM observability & tracing
Evaluation & testing
Safety, guardrails & red-teaming
Workflow automation
Durable execution & orchestration
API & developer workflow tools
Productivity & collaboration
Internal tools & databases
CRM, ERP, support & identity
Design, CMS & commerce
Open web / browser / search