Side-by-side comparison of Step-3.7-Flash (StepFun) and Talkie-1930-13B-IT (Talkie-LM (research)) for self-hosted deployment of the open-weight model. Step-3.7-Flash is rated conditional; Talkie-1930-13B-IT is conditional. They part ways on training data: Step-3.7-Flash is "Undisclosed", Talkie-1930-13B-IT is "Documented".
| Field | ||
|---|---|---|
| Summary | ||
| Verdict | 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. | Conditional Per the published model card, Talkie-1930-13B-IT is an Apache 2.0 instruction-tuned 13B model trained exclusively on pre-1931 English text (260B tokens, sourced from public-domain reference works). The training-data transparency is unusually clean for AI Act Article 53 purposes; the limits are vendor jurisdiction (a US-affiliated research collaboration with no published EU DPA) and the deliberate vintage corpus, which makes the model unsuitable for any task requiring post-1931 factual knowledge. |
| Last reviewed | 2026-05-30 | 2026-04-28 |
| Open-weight | ||
| Licence | Apache 2.0 | Apache 2.0 |
| Commercial use | Unrestricted | Unrestricted |
| Training data | Undisclosed | Documented |
| Origin | China (Shanghai) | US (research) |
| Performance & pricing? | ||
| Quality index | — | — |
| Speed | — | — |
| Blended price | — | — |
| Context window | — | — |
| Evidence | ||
| Sources | ||
No overlapping sources between the two entries.