Cerebras has expanded its public endpoints with the addition of the Qwen 3.8 27B model, now available for use on both free trial and pay-as-you-go tiers. This model is part of Cerebras' commitment to providing a range of models for various use cases, all of which can be accessed through their public API.

Browse all models available on Cerebras public endpoints to find the one that best suits your needs. For those new to the platform, a Quickstart guide is available to help make the first API call, and a model selection guide can aid in choosing the right model by use case. Each model listed comes with full specifications, capabilities, and per-tier limits for transparency.

Cerebras also hosts a variety of open-source models from the community, though it's noted that these are the original, unpruned versions. The company conducts research on pruning techniques, such as Router-weighted Expert Activation Pruning (REAP), but these pruned models are shared with the research community on Hugging Face and are not available through the public API.

The company uses selective weight-only quantization during storage to preserve maximal quality, storing weights in partial 16-bit, 8-bit, or 4-bit, in line with industry standards. Quality-sensitive layers are stored at full precision, with dequantization on the fly to ensure operations are done in high precision. Activations, attention, and kv cache remain in full precision and unquantized.

Users can be assured that Cerebras does not alter model architectures without notice. The original models are served for all existing endpoints without modification. If the company explores additional compression techniques in the future, these would be offered as separate endpoints, ensuring transparency and choice for users.

For those interested in Cerebras' REAP pruned models, they are available on Hugging Face for research and experimentation purposes. This collection demonstrates the company's pruning research but is not part of the production API.

Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.