Fastino's GLiDE adds adaptive thinking to a hosted decision model
Fastino launched GLiDE (“Generalized Lightweight Decision Engine”) on 1 October 2026 at 15:27 BRT, in posts by @fastinoAI and @george_onx, with a blog post dated 30 September by Sahibzada Allahyar and Mary Newhauser.
What it does. GLiDE first produces a fast probability distribution over the options. When the leading answer is uncertain, it reasons further and folds that into the final probabilities. The reasoning tokens show up as output_tokens and are free. The API sits at POST https://api.fastino.ai/v1/systemone (model fastino/glide) and takes Noul, Choice (up to 255 options) and Score with up to 40k tokens. Fastino’s docs say it is not a drop-in for Jev: score returns an integer index plus expected_level, and confidence is the gap between the top two options. Price: $0.30 / M input tokens. Weights, size and base model are not disclosed.
The numbers (UnverifiedClaim). 64.81 on Decision Index 0.2.1 vs Jev’s published 57.91, from Fastino’s own complete run with the official scorer; ahead in all five areas and on 31 of 38 benchmarks; Knowledge and Reasoning 62.9 vs 51.4; CLadder 88.7 vs 72.6; CRUXEval 92.6 vs 73.0. @m_newhaus adds +6.9 on Humanity’s Last Exam. The public board has paused 0.2.1 submissions, so George’s “Ranks #1 on the Decision Index” is not something the board shows. Cloudflare claimed the lead for Clef the same day, also self-run. No latency numbers are published.
Other notes. The launch post says “inference and training”, but the docs list no fine-tuned GLiDE deployments. “First thinking decision model” is marketing: TypeLLM has sold a thinking mode for decisions since 29/09.
How we list it. Catalog, 4.9 maturity / 6.0 capability / 13 adoption.