Open-Source Model Selector
Choosing an open model is really four decisions, not one: what you are building, how sensitive the data is, how much infrastructure you can run, and what you can spend. This free tool asks those four questions and returns one or two open models that fit, with the license class spelled out and an honest note on when a closed frontier model would still serve you better. It is a rule engine over our verified comparison table, so every answer traces back to a model we have actually checked.
Read this first: the word "open" hides three legal realities. OSI open source (Apache 2.0, MIT) lets you use, modify, and redistribute commercially. Open-weight means the weights download freely but the license restricts use, as with the Llama 4 Community License. Source-available, non-free licenses block commercial deployment. The selector labels which class every recommendation falls into, because for a business that distinction is the whole decision.
Run the Selector
Answer the four questions and the tool returns a primary pick and a secondary option, each with its license and the caveats your answers imply. Nothing is stored; this free version runs entirely in your browser.
Put governance around how your team uses AI. The AI Acceptable Use Policy: a deploy-ready template that sets the rules for AI use.
Your purchase helps keep our hubs free to read.
How It Reads Your Answers
The logic is deliberately simple and transparent, so you can audit it. Your use case selects a base model family. Your infrastructure maturity and budget then choose between a heavier self-hosted variant and a lighter one: managed-only infrastructure or a minimal budget steers you to the model you can run without a cluster. Your data sensitivity adds the residency and licensing caveats, which is where the China-hosted DeepSeek API, the proprietary Qwen Max API, and the EU restriction in the Llama license all get flagged.
- General chat: Mistral Large 3 (Apache 2.0) when you can self-host, Gemma 3 27B when you cannot.
- Coding and agents: Qwen3-Coder or the lighter Qwen3.6-35B-A3B, both Apache 2.0 open weights.
- Long-context: DeepSeek V4-Pro or the lighter V4-Flash, both MIT, with Llama 4 Scout noted when you truly need beyond 1M tokens.
- On-device / edge: Phi-4 Reasoning (MIT) or Gemma 4 edge variants (Apache 2.0).
- Multilingual: Mistral Large 3 or the lighter Ministral 3, both Apache 2.0 with European data residency.
When a Frontier Model Is Still the Better Call
Open weights hand you the operational burden along with the control. If you need the absolute top-end reasoning or multimodal ceiling, a zero-ops managed API with an enterprise SLA, or you cannot staff self-hosting at all, a managed frontier model such as GPT-5.5, Claude Opus 4.8, or Gemini 3 is the right answer. The sharpest case is regulated data on managed-only infrastructure: the cleanest open path needs self-hosting you cannot run, so a compliant, region-locked frontier model can be safer than a China-hosted open API. The tool surfaces this note with every result rather than pretending open is always the answer.
The Models It Draws From
These six are the only models the selector can output. Each carries a license we have verified against primary sources. For the full profiles, trade-offs, and the five more models on the watch list, read the Best Open-Source AI Models comparison.
| Model | Maker | License class | Best fit |
|---|---|---|---|
| Llama 4 (Maverick / Scout) | Meta | Open-weight (not OSI) | Ultra-long context, multimodal |
| DeepSeek V4 (Pro / Flash) | DeepSeek | MIT (OSI open source) | Cheap long-context reasoning |
| Qwen open line (Coder / 35B-A3B) | Alibaba | Apache 2.0 (OSI open source) | Agentic coding |
| Mistral Large 3 / Ministral 3 | Mistral AI | Apache 2.0 (OSI open source) | Multilingual, EU residency |
| Gemma 4 / Gemma 3 27B | Google DeepMind | Apache 2.0 / source-available | Single-accelerator multimodal |
| Phi-4 Reasoning | Microsoft | MIT (OSI open source) | On-device reasoning |
Models verified 2026-06-30. Licenses and context windows move fast; confirm on each model card before you build.
Test Your Open-Source Licensing Knowledge
The selector is only as useful as your grasp of what "open" really means. Three tiers, from a quick check to a procurement-grade scenario.