CUDOS

SingularityNET

Singularity

Singularity Compute, which runs on <a href="https://x.com/SingularityNET" rel="noopener noreferrer nofollow">@SingularityNET</a>'s tech, has released Super Decider, a model that anyone can download and use for free under an open license. Instead of writing text, it picks from options a developer sets, such as which team should handle a support ticket, and gives each option a probability in about 0.3 seconds.
The model is live on ASI:Cloud (<a href="https://x.com/CUDOS_" rel="noopener noreferrer nofollow">@CUDOS_</a>). In the company's own testing, it scored 85.6 on NVIDIA's When2Call benchmark, ahead of TypeSafe's Jev and Cloudflare's Clef, but 1.6 points behind Clef on a tool-selection test. The figures come from the developer and have not been independently verified.
<h2 id="how-super-decider-works">How Super Decider works</h2>
According to the project's <a href="https://github.com/Singularity-Compute/super-decider" rel="noopener noreferrer nofollow">GitHub repository</a>, Super Decider answers typed questions about a state, covering choices, yes/no questions and scores, each with a calibrated probability distribution returned in a single call. The repository describes it as a fine-tune of Google's DiffusionGemma 26B-A4B, a discrete-diffusion mixture-of-experts model with 4 billion active parameters. It reads every answer slot in one denoising step and runs on a single NVIDIA L40S GPU.
The same page says the API follows the Jev / TypeSafe "System One" request format and is also compatible with OpenAI-style chat completions. The code is released under the Apache-2.0 licence. The developers state that no benchmark test items were used for training.
For developers, the appeal is a structured output rather than free-form text. A classifier-style model that reports its own confidence can be used to route support tickets, triage requests or gate automated actions, with low-confidence cases passed to a human.
<h2 id="what-the-benchmarks-measure">What the benchmarks measure</h2>
When2Call is an NVIDIA benchmark. Its <a href="https://aclanthology.org/2025.naacl-long.174/" rel="noopener noreferrer nofollow">peer-reviewed paper</a> says it evaluates tool-calling decision-making: when to generate a tool call, when to ask follow-up questions and when to admit a question cannot be answered with the tools provided. That focus on the decision, rather than the accuracy of the call itself, is why it suits a model built to choose between options.
The result sits within a broader push by the ASI Alliance into AI infrastructure. <a href="https://www.singularitycompute.com/" rel="noopener noreferrer nofollow">Singularity Compute</a> describes itself as part of SingularityNET and the ASI Alliance, and delivers inference through the ASI:Cloud platform. The alliance links Fetch.ai, SingularityNET and CUDOS under a shared token, according to <a href="https://crypto.com/en/university/what-is-the-artificial-superintelligence-alliance" rel="noopener noreferrer nofollow">Crypto.com Academy</a>.
Because the benchmark results are self-reported and the comparison with Clef on tool selection is narrowly negative, developers may want to test the model on their own workloads before adopting it.
Sources
<a href="https://github.com/Singularity-Compute/super-decider" rel="noopener noreferrer nofollow">Super Decider repository, Singularity Compute (GitHub)</a> <a href="https://aclanthology.org/2025.naacl-long.174/" rel="noopener noreferrer nofollow">When2Call: When (not) to Call Tools, ACL Anthology</a> <a href="https://crypto.com/en/university/what-is-the-artificial-superintelligence-alliance" rel="noopener noreferrer nofollow">What Is the Artificial Superintelligence Alliance?, Crypto.com</a>

ASI Alliance’s new open AI picks an answer and says how sure it is

Table of Contents

BSCN

Other articles published on Oct 23, 2026

Join our free newsletter for daily crypto updates!