Shortlist snapshot
| Model | Architecture / size | License | OWM hardware class | Provider origin |
|---|---|---|---|---|
| gpt-oss-20bOpenAI | 20B · compact reasoning | Apache 2.0 | 17–32B · high-memory workstation | United States |
| Qwen3-32BQwen / Alibaba | 32B · dense | Apache 2.0 | 17–32B · high-memory workstation | China |
| Qwen3-Coder-30B-A3B-InstructQwen / Alibaba | 30B / ~3B active · coding MoE | Apache 2.0 | 17–32B · high-memory workstation | China |
| Devstral Small 2 24B Instruct 2512Mistral AI | 24B · coding agent model | Apache 2.0 | 17–32B · high-memory workstation | EU provider |
| Mistral Small 4 119B A6BMistral AI | 119B / 6.5B active · multimodal MoE | Apache 2.0 | Model-specific · large / specialized | EU provider |
| Granite 4.2 8BIBM | 8B · 128K enterprise reasoning | Apache 2.0 | ≤8B · consumer/local | United States |
| OLMo 3 32BAi2 | 32B · fully open research stack | Apache 2.0 | 17–32B · high-memory workstation | United States |
| SmolLM3 3BHugging Face | 3B · hybrid reasoning | Apache 2.0 | ≤8B · consumer/local | United States |
What should drive the decision?
A license label should be attached to a specific checkpoint, not inferred from a developer or model family. Publishers can use different terms across releases, and downstream quantizations can add their own metadata or packaging considerations.
Commercial suitability also depends on the application: redistribution, hosted inference, fine-tuning, generated-output policy, third-party datasets and model-specific use restrictions can create separate obligations.
OWM therefore exposes license, provider origin and EU deployment as separate fields. Apache 2.0 is a licensing fact; it is not a GDPR, AI Act, security or data-residency certification.
Models to evaluate
gpt-oss-20b
Reasoning, local and edge-class inference
OpenAI reasoning model; Apache 2.0 plus model-specific usage policy context.
Qwen3-32B
Reasoning, multilingual, general-purpose
General-purpose dense Qwen3 checkpoint under Apache 2.0.
Qwen3-Coder-30B-A3B-Instruct
Coding agents, repository work, tool use and software engineering
Agentic coding checkpoint under Apache 2.0.
Devstral Small 2 24B Instruct 2512
Software engineering, repository-scale coding and agentic development
EU-provider coding model under Apache 2.0.
Mistral Small 4 119B A6B
Instruction, reasoning, coding, agents and vision
Large multimodal MoE from Mistral AI under Apache 2.0.
Granite 4.2 8B
Enterprise assistants, reasoning, tool use and RAG
Compact IBM enterprise/RAG-oriented Apache model.
OLMo 3 32B
Open research, reproducibility, instruction and reasoning variants
Ai2 model with a strong open-research and reproducibility orientation.
SmolLM3 3B
Compact reasoning, agents and local inference
Compact Hugging Face model for low-footprint local use.
Hardware and runtime reality
Weight-only estimates are a starting point. Add KV cache, runtime workspaces, multimodal components, batching and concurrency headroom. For local inference, validate the exact quantized artifact. For server inference, measure time to first token, throughput and peak memory at target concurrency.
Long context can make an otherwise comfortable model exceed the practical memory budget. Test the longest realistic prompt and generation, not only a short loading test.
License, provider origin and Europe
Review the exact checkpoint license and any separate usage terms. Provider origin is supply-chain metadata, not an inference-location claim. If EU/EEA residency matters, map inference, RAG, embeddings, logs, telemetry, backups and subprocessors.