Deepseek, a Chinese company specializing in language models, is testing a new model whose architectural details are currently unknown.

Preliminary benchmarks

The first tests focus on the model's ability to understand long texts. The results, published on Reddit, show a series of evaluations on different indices, with context windows of 128,000 and 256,000 tokens. Some tests were passed, others were not. The model name used in the benchmarks is a placeholder.

For those evaluating on-premise deployments, there are trade-offs to consider. AI-RADAR offers analytical frameworks on /llm-onpremise to evaluate these aspects.