Launching today
GLiDE
The First Thinking Decision Model
2 followers
The First Thinking Decision Model
2 followers
GLiDE is Fastino's thinking decision model. Give it app state and a set of options, and it returns one chosen action, a confidence score, and a probability for every option, so software can act without parsing text. It answers fast when confident and reasons further when unsure. It scores 64.81 on Decision Index 0.2.1 vs Jev's 57.91, with a 40k token context window. Available via the Fastino API.
GLiDE is Fastino Labs' thinking decision model, built for difficult, structured decisions.
Problem: When an AI agent has to choose its next step, like which tool to call or whether to run a risky command, the usual move is to ask a big model and parse a free-text answer. Traditional decision models score the options in a single fixed pass, which works for routine classification but struggles when a decision depends on arithmetic, code tracing, dates, conflicting evidence, or several related facts.
Solution: GLiDE takes your application state and a predefined set of actions, then returns one selected action, a confidence score, and a probability for every option. Because the output is structured, you can execute the action, set confidence thresholds, ask a human for approval, or escalate uncertain cases with ordinary application logic, with no text to parse or validate.
What makes it different: Adaptive thinking. GLiDE makes a fast first pass, and when it's confident it answers right away. When it isn't, it spends extra reasoning and folds that into its final probabilities. Easy decisions stay fast, and hard ones get more compute.
On Decision Index 0.2.1 it scores 64.81 vs Jev's published 57.91 (+6.90 skill points), leading in all five evaluation areas and 31 of 38 benchmarks. The biggest gaps are in reasoning-heavy tasks: Knowledge and Reasoning (62.9 vs 51.4), Tools and Automation (83.5 vs 75.1), CRUXEval (92.6% vs 73.0%) and CLadder (88.7% vs 72.6%). These come from Fastino's own run with the official scorer.
Key features:
Returns a selected action, a confidence score, and probabilities for every option
Adaptive thinking: fast when confident, deeper reasoning when uncertain
Handles decisions across hundreds of tools or possible paths
40k token context window for policies, contracts, support histories and runbooks
One API format for any app that combines context with a defined set of choices
Available now on the Fastino API
Who it's for: Developers building agents and AI workflows that need to pick actions, route requests, or verify outputs. Fastino's examples include navigating large tool sets, deciding whether to execute, sandbox, request approval or block a high-risk agent action, diagnosing production incidents, applying complex contracts and policies, and checking reasoning-heavy model output.
Try it via the Fastino API, see pricing and the request schema in the docs, or read the launch post.