@runedotgl is live on Fireworks serverless.
It's a System One decision model: 26B parameters of pretrained knowledge, only 4B active per token. It doesn't reason or generate text. It reads a document once and answers up to 256 questions (multiple choice, true/false, graded scale) in a single pass, each with a calibrated probability.
→ Median latency under 150ms, measured server-side
→ 262K context, text only
→ $0.042/1M input, $0.008 cached, $0 output
→ Serverless or on-demand deployments
Try it out → http://fireworks.ai/models/fireworks/rune