DeepSeek v4.1 Flash at 378 tok/s 99.7% Cache hit rate (runinfra.ai)

DeepSeek has announced the release of version 4.1 Flash, a large language model (LLM) available through RunInfra Model APIs, offering impressive performance at a speed of 378 tokens per second and a remarkable 99.7% cache hit rate. For a limited time, the model is priced at just $0.10 per million input tokens, with discounted rates for cached input and output tokens until October 13, 2026. This version supports OpenAI-compatible chat completions and Anthropic-compatible messages, making it versatile for various applications. The introduction of DeepSeek V4.1 Flash is significant for the AI/ML community as it provides enhanced capabilities in handling large conversation contexts, with a context window of over a million tokens. The model's automatic prefix caching improves efficiency by quickly retrieving relevant conversation history, though cache retention is based on available memory. These features position DeepSeek V4.1 Flash as a competitive player in the LLM market, especially for developers looking for cost-effective and efficient solutions in real-time communication and data processing.