Bittensor Produces the World’s Top-Ranked Decision Model With REAX 1.3

Bittensor has produced the highest-scoring decision model on Hugging Face’s Decision Index. REAX 1.3, developed through REAX (SN92), achieved a score of 74.0, leading a field of 275 models evaluated for their ability to make fast, structured decisions.

The model outscored Fastino’s GLiDE 27B, which recorded 73.4, and PelaAI’s KnowLine 31B Gen1.5 at 71.9. The result places REAX 1.3 at the top of Decision Index 0.4.1, a benchmark that tests how well AI models make judgments across different tasks and unfamiliar situations.

What makes the achievement particularly remarkable is how quickly it happened. REAX launched on Bittensor in October 2026, and within weeks, its open competition had helped produce a model capable of competing with leading AI research teams.

How Bittensor Achieved This

REAX (SN92) operates an open competition where miners from around the world develop and improve AI decision models. Their submissions are evaluated for performance, with rewards designed to encourage measurable improvements.

For REAX 1.3, the team selected two promising models developed by competing miners and combined them with other high-performing open models. One builds on the Gemma family, while the other uses the Qwen family. Both run together, and their predictions are combined to produce the final result.

This approach allowed REAX to build on the work of multiple independent contributors without training an entirely new model from scratch.

It also demonstrates one of Bittensor’s biggest strengths. By opening AI development to talent worldwide and rewarding measurable contributions, decentralized networks can accelerate experimentation and bring together expertise that a single company might struggle to assemble.

Matt Mastracci, the person behind KindlingAI, congratulated Florian and the REAX team for assembling talent from across the internet to build an open-weight competitor to Jev. He also highlighted the speed of development, noting that System One-style models only emerged a little over a month ago.

What REAX 1.3 Does and Why It Matters

Decision models help software make quick judgments without generating lengthy text responses.

Imagine a company receiving thousands of customer complaints. An AI system needs to determine which department should handle each complaint, whether it is urgent, and whether human attention is required.

A decision model can evaluate the information and return probabilities for the available options. This makes it useful for customer support, fraud detection, risk assessment, and AI agents that need to choose what action to take next.

REAX 1.3 performs these judgments without generating text, making its outputs easier for other software to use directly. Its model weights are publicly available on Hugging Face under the Apache 2.0 license, and developers can also access it through a compatible API.

A Major Milestone for Bittensor

REAX 1.3 is an example of what decentralized AI development can achieve in a short period. Within weeks, an open network of miners contributed to a model that reached the top of an international AI benchmark.

The achievement also highlights a different approach to AI research. Bittensor allows talent from anywhere in the world to compete, contribute improvements, and earn rewards based on measurable performance.

With REAX 1.3, that approach has delivered a world-leading result in a growing category of AI models, strengthening the case for decentralized networks as a serious force in AI development.