raullenchai/Rapid-MLX v0.6.20: Apple Silicon AI Engine Claims 4.2x Faster Than Ollama, Released Today

GitHub·medium signal

Rapid-MLX released v0.6.20 today (May 7) — an Apache 2.0 local inference engine for Apple Silicon claiming 4.2x faster performance than Ollama with 0.08s cached TTFT. The OpenAI-compatible API server supports tool calling with 17 parser formats, persistent prompt caching, reasoning separation, cloud routing for large contexts, and multimodal input. Supports Qwen3.5, Gemma 4, DeepSeek V4 Flash, and Llama 3.3 across Mac configurations from 16GB Air to 256GB Studio Ultra. 1.8K stars.