Side-by-side comparison of LFM2.5-8B-A1B (Liquid AI) and Step-3.7-Flash (StepFun) for self-hosted deployment of the open-weight model. LFM2.5-8B-A1B is rated conditional; Step-3.7-Flash is conditional. They part ways on licence: LFM2.5-8B-A1B is "LFM Open License 1.0", Step-3.7-Flash is "Apache 2.0".
| Field | ||
|---|---|---|
| Summary | ||
| Verdict | Conditional Per the published LFM Open License v1.0, Liquid AI's terms are Apache-2.0-derived but cap free commercial use at USD 10M annual company revenue — above that threshold deployers must negotiate a paid licence with [email protected]. The model itself is a strong 8.3B / 1.5B-active MoE built for CPU and edge inference, but the 38-trillion-token training corpus is undisclosed and Liquid AI is a US-based MIT spinoff, so EU regulated buyers should run their own DPIA and document the data-transfer story alongside any commercial-tier negotiation. | Conditional Per the published Apache 2.0 LICENSE, the Step-3.7-Flash weights ship without commercial restriction — a 198B / 11B-active vision-language MoE with a 256k-token context aimed at tool-heavy and agentic workflows. The remaining EU-readiness gaps are the entirely undisclosed training corpus and StepFun's Shanghai-based vendor jurisdiction; deploy on self-managed EU infrastructure and document the Art. 50 transparency story for any synthetic or biometric output. |
| Last reviewed | 2026-05-30 | 2026-05-30 |
| Open-weight | ||
| Licence | LFM Open License 1.0 | Apache 2.0 |
| Commercial use | Free under $10M revenue | Unrestricted |
| Training data | Undisclosed | Undisclosed |
| Origin | USA (Boston) | China (Shanghai) |
| Performance & pricing? | ||
| Quality index | — | — |
| Speed | — | — |
| Blended price | — | — |
| Context window | — | — |
| Evidence | ||
| Sources | ||
No overlapping sources between the two entries.