Apertus 70B is a large-scale open language model developed by the Swiss AI Initiative as part of a fully-open multilingual family that also includes an 8B variant, with the Instruct release dated to September 2025. Independent community distributions confirm the 70-billion-parameter design and describe the model as built to push the boundaries of fully transparent, multilingual language modeling, making it well suited for research, evaluation, and applications where openness of weights and training intent matter. The open-weights framing means practitioners can inspect, fine-tune, or self-host the model, and community re-quantized builds (such as a 44 GB q4_k_m Instruct artifact on Ollama) demonstrate the practical availability of the weights beyond the original release channels.

For practical use, Apertus 70B is positioned as a general-purpose multilingual text model rather than a narrow specialist, with deployment options spanning API hosting and self-hosting of the open weights. Third-party provider benchmarking currently shows the model served through Public AI at a blended price of $1.03 per million tokens, while per-provider speed and latency benchmarks are not yet published, so users should evaluate throughput against their own workloads. The combination of an explicit multilingual objective, open-weights availability, and a flagship 70B scale makes it a natural fit for teams that want a transparent foundation model for experimentation, multilingual assistants, or downstream fine-tuning without locking into a single closed provider.