AI Tools Index / Open models / Compare
comparison · 15 models · same five criteria, same maximum
Every model score, side by side.
Read the columns that matter for your constraint. A firm with a hard commercial-licence requirement should start with the licence column; a firm whose supervision model depends on citation discipline should start with evidence. The total is a summary, not a verdict.
| Model | Total/100 | Tier | Licence & commercial freedom25% | Legal-task calibration25% | Evidence & abstention discipline20% | Context & retrieval behaviour15% | Deployment reality15% | Categories |
|---|---|---|---|---|---|---|---|---|
| QwenAlibaba Cloud | 84 | B | 23/25 | 20/25 | 14/20 | 14/15 | 13/15 | Frontier-class open weights · Efficient and on-premise |
| gpt-ossOpenAI | 84 | B | 23/25 | 20/25 | 15/20 | 12/15 | 14/15 | Frontier-class open weights · Efficient and on-premise |
| MistralMistral AI | 83 | B | 21/25 | 20/25 | 15/20 | 14/15 | 13/15 | Frontier-class open weights · Efficient and on-premise |
| GraniteIBM | 82 | B | 23/25 | 18/25 | 14/20 | 12/15 | 15/15 | Efficient and on-premise |
| GemmaGoogle DeepMind | 81 | B | 22/25 | 19/25 | 14/20 | 13/15 | 13/15 | Efficient and on-premise |
| KimiMoonshot AI | 80 | B | 20/25 | 19/25 | 15/20 | 15/15 | 11/15 | Frontier-class open weights |
| GLMZhipu AI | 79 | B | 22/25 | 19/25 | 14/20 | 13/15 | 11/15 | Frontier-class open weights |
| LlamaMeta | 78 | B | 17/25 | 21/25 | 15/20 | 13/15 | 12/15 | Frontier-class open weights |
| NemotronNVIDIA | 77 | B | 15/25 | 21/25 | 15/20 | 13/15 | 13/15 | Frontier-class open weights |
| DeepSeekDeepSeek | 76 | B | 22/25 | 19/25 | 13/20 | 12/15 | 10/15 | Frontier-class open weights |
| OLMoAllen Institute for AI | 75 | B | 22/25 | 14/25 | 15/20 | 10/15 | 14/15 | Fully transparent |
| ApertusSwiss AI Initiative / ETH Zürich and EPFL | 73 | B | 17/25 | 16/25 | 15/20 | 12/15 | 13/15 | Fully transparent |
| PhiMicrosoft | 71 | C | 24/25 | 15/25 | 11/20 | 9/15 | 12/15 | Efficient and on-premise |
| CommandCohere | 71 | C | 18/25 | 17/25 | 13/20 | 12/15 | 11/15 | Frontier-class open weights |
| Legal-domain open modelsVarious academic and domain groups | 60 | C | 16/25 | 13/25 | 12/20 | 8/15 | 11/15 | Legal-domain and legal-tuned |
before you choose on score alone
Deployment reality varies more between firms than between models: a 12-fee-earner practice and a 120-fee-earner firm are not choosing the same hardware, and the evaluation burden is the same either way. Start with the readiness assessment if the shortlist is still open.