Magnitude, a Y Combinator S25 startup, published an open source inference engine on 30 September that it says runs open models up to 2x faster than llama.cpp on Apple Silicon, NVIDIA, AMD or a plain CPU, with no token costs. The same day, a price index covering 81 public records put the median listed output price at $6 per 1M tokens against $1.32 for input.
AI & models30 September 20266 min readSources 10
An open source inference engine that tunes its CPU kernels on your own device launched on 30 September, claiming decode speeds up to 92% faster than llama.cpp on Metal, as new research and pricing data point to a widening split in the cost of running AI.
AI & models30 September 20266 min readSources 9
Nvidia published Kumo Tabular, an open foundation model for tabular prediction, in a Hugging Face post dated 29 September, while Fermion Research released the 164 MB speech model Phonon-2 a day later. Both arrive as the FTC opens an industry-wide investigation into AI labs and OpenAI holds back its own IPO.
AI & models30 September 20263 min readSources 6
Two research teams published new inference-engine results on 30 September, claiming AI-SQL queries can run more than ten times faster than a vLLM baseline on the same hardware.
AI & models30 September 20264 min readSources 7
Two open source inference projects published benchmarks on 30 September claiming large cuts in the cost of running AI workloads, as a separate pricing index put the median output token at $6 per million.
AI & models30 September 20263 min readSources 5
Two engineering teams published separate claims on 30 September that the cost of running open models can be cut by rewriting how inference engines handle hardware and query structure, not by waiting for cheaper GPUs.
AI & models30 September 20263 min readSources 6