Rune launches 26B-A4B 'System One' decision model: no text generation, answers in one pass
sophiamyang · x · 2026-10-30
Rune's 26B-A4B is now live on Fireworks serverless as a 'System One decision model': 26.5B-param MoE with 4B active per token that never reasons or generates text.
- Reads a document once and answers up to 256 typed questions (multiple choice, true/false, graded scale) in a single pass, each with calibrated probabilities
- Median latency under 150ms server-side; 262K context, text only
- Zero output tokens means input-only pricing: $0.042/1M input, $0.008 cached, $0 output
- Available serverless or on-demand; GGUF weights are on Hugging Face
A complement to generative LLMs, aimed at classification and bulk-judgment workloads.
More from Infra
- Nvidia to Guarantee Over $5B in Debt for Nscale's 460MW Anthropic Data Center — rohanpaul_ai · 2026-10-30
- Nvidia reportedly to guarantee $5B+ construction debt for Nscale's 460MW Anthropic data center — rohanpaul_ai · 2026-10-30
- Even 2020-era GPUs keep gaining rental value, undercutting the AI bubble thesis — JFPuget · 2026-10-30
- Intel earnings call: server CPU TAM seen 5x to $200B+, chip shortage into 2028 — ryanshrout · 2026-10-30
- Lightweight FastAPI proxy gateway adds auth for locally hosted LLMs in OpenCode — Frizzy-MacDrizzle · 2026-10-30
- Nvidia reportedly in talks to backstop >$5B on Nscale's 460 MW site, locking in Anthropic's $45B Vera Rubin deal — ns123abc · 2026-10-30