#United States #San Francisco #AI Model #Fastino Labs #Glide

Fastino Labs Unveils GLiDE: A Breakthrough in Decision-Making Models

On October 1, 2026, Fastino Labs, known for its pioneering work in applied AI, introduced GLiDE (Generalized Lightweight Decision Engine), a groundbreaking adaptive decision-making model that is set to revolutionize how AI systems understand and process choices. This model has taken the lead in the Decision Index, outperforming the long-standing top model, Jev, by an impressive 6.9 points. This leap in performance places GLiDE firmly at the forefront of AI decision-making technology, representing a substantial advancement in the field.

Performance Metrics

In a recent evaluation using the Decision Index version 0.2.1, GLiDE scored an impressive 64.81 while Jev managed 57.91. These results were based on Fastino's extensive testing across a robust dataset comprising 155,390 requests. Moreover, GLiDE didn't just win; it exceeded expectations in all five evaluation areas and claimed victories in 31 out of 38 specific performance benchmarks.

The standout improvements for GLiDE were evident in categories such as Knowledge and Reasoning (+11.5), Tools and Automation (+8.4), and Arts and Human Taste (+8.3). For instance, in Knowledge and Reasoning, GLiDE triumphed over Jev in 7 of the 9 benchmarks tested. Notable wins were on the most challenging tasks, where GLiDE achieved remarkable scores: 32 points ahead in CLadder (which assesses cause and effect reasoning), 31 points ahead in CRUXEval (focused on code output predictions), and 24 points ahead in GSM8K (concerned with grade-school math problems).

Addressing Challenges in AI Decision-Making

As AI begins to play significant roles in real-world applications, the complexity of decision-making has become increasingly apparent. Rather than just generating content, AI systems must now determine the next steps, decide which tools to utilize, seek approvals, and even know when to halt actions. Traditional systems often resort to querying a broader general-purpose model to make these decisions, which can be inefficient, costly, and unreliable in operational settings. In contrast, GLiDE introduces a more structured and reliable framework for making decisions.

Advantages of GLiDE

GLiDE utilizes a structured approach: it evaluates the current state of an application, considers a predefined set of options, and returns a specific action along with a confidence score for that action as well as probability figures for all available choices. By structuring its output, developers can immediately act on these results, applying confidence thresholds or transferring uncertain cases to human operators, thereby eliminating the need to process free-form text.

Moreover, GLiDE embodies an adaptive thinking process. Initial decision paths are swift, with many decisions resolved quickly on a more straightforward basis. When faced with more complex queries, GLiDE engages in deeper reasoning, gradually integrating that thought process into the final outputs. Testing revealed that approximately two-thirds of requests could be addressed rapidly, allowing intricate and challenging decisions to benefit from additional computational support.

Targeting Difficult Decision-Making Scenarios

GLiDE shows substantial gains in areas demanding intensive reasoning and tool usage, leading by considerable margins over Jev. For instance, it achieved an accuracy score of 92.6% on CRUXEval, surpassing Jev's 73.0%, and on CLadder, it garnered 88.7% compared to Jev's 72.6%. This advanced contextual understanding, supported by a 40,000 token context window, positions GLiDE as an ideal solution for handling extensive policies, contracts, and operational histories.

The Vision for AI Decision-Making

George Hurn-Maloney, CEO and co-founder of Fastino Labs, stated, "We envision decision-making evolving into its own layer within the AI architecture, similar to how retrieval processes have developed. There's no need to reach out to a cutting-edge model every time an AI agent must make a choice. With our GLiNER model already surpassing 50 million downloads, GLiDE stands as our most proficient decision-maker to date, capable of swiftly managing routine decisions while applying cognitive depth to tougher challenges."

Practical Applications of GLiDE

GLiDE's functionalities can be adapted across various sectors and tasks, ranging from standard classification and routing to those requiring advanced reasoning. Examples include:

• - Navigating complex action graphs: Selecting the next steps from numerous tools and potential paths, while considering prerequisites and possible results.
• - Managing high-risk actions: Determining actions based on system states, permissions, and organizational guidelines.
• - Diagnosing operational incidents: Making informed choices based on alerts, logs, and recent changes in system health.
• - Applying intricate contracts and policies: Identifying whether claims should be approved, rejected, or modified.
• - Validating complex AI outputs: Assessing responses that depend on various reasoning processes before final acceptance.
• - Resolving operational constraints: Making a selection that adheres to predefined rules, resource limits, and other essential criteria.

Availability

GLiDE is officially available now via the Fastino API, recognized by the model ID fastino/GLiDE. Interested users can explore its potential and pricing options at the official Fastino documentation site.

About Fastino Labs

Founded in 2024, Fastino Labs is a cutting-edge applied AI research organization focusing on the development of lightweight open models and infrastructure to enhance their continuous improvement in practical applications. As the inventor of the GLiNER model family, Fastino Labs contributes significantly to advancing AI's capabilities, backed by major investors such as Khosla Ventures, Insight Partners, M12, and NEA. For more information, visit fastino.ai.

Fastino Labs continues to pave the way for innovative AI solutions that prioritize efficiency, reliability, and extensive adaptability in decision-making.