Qubic Tests GPU-Free Evolutionary AI on ARC-AGI-3
Qubic’s research team used evolutionary architecture search to train Neuraxon agents on CPUs, achieving a best Kaggle score of 0.41.
Qubic’s research team used evolutionary architecture search to train Neuraxon agents on CPUs, achieving a best Kaggle score of 0.41. This is an original Token Times brief based on reporting by Cointribune, with live market data added, as of Tue, 17 Nov 2026 02:45:02 UTC.
The experiment shows that automated architecture search can produce competitive benchmark results without GPU hardware or pretrained models. It also expands the practical scope for CPU-based AI research and experimentation.
- More than 110 configurations and millions of evolved brains were tested.
- The experiments used about 500 CPU cores, with no GPU or pretrained model.
- Neuraxon’s best offline score was 0.38; its latest Kaggle result was 0.41.
- Crypto markets are broadly higher, while Fear & Greed remains at extreme fear.
Qubic’s scientific team applied evolutionary neural architecture search to Neuraxon agents competing in ARC-AGI-3, the interactive benchmark associated with ARC Prize 2026. According to Cointribune, the research tested more than 110 configurations and millions of evolved brains using approximately 500 CPU cores, without a GPU or pretrained model.
Rather than manually designing each agent’s architecture, the team allowed an evolutionary process to search for effective brain designs. Cointribune reported that the best result in offline testing reached 0.38, while a later Kaggle submission improved the score to 0.41.
The account is based on Neuraxon Intelligence Academy Vol. 14 and is the only supplied source, so the results and experimental approach have not been independently corroborated in this coverage. It also says the underlying code and a 603,054-row dataset can be downloaded, while Neuraxon 3.0 is expected to continue the search process with broader or updated objectives.
The development is not tied to a specific cryptocurrency, but it arrives against a positive yet sentiment-constrained market backdrop. Total crypto market capitalization stands at $2.26 trillion,
Bitcoin is up 2.99% over 24 hours, and the Fear & Greed Index is at 10, indicating extreme fear. The immediate figures to watch are whether Neuraxon’s next evolutionary runs improve beyond the reported 0.41 Kaggle score and what additional objectives the 3.0 release introduces.
Qubic Tests GPU-Free Evolutionary AI on ARC-AGI-3 — questions & answers
What score did Neuraxon achieve on ARC-AGI-3?
The best offline score was 0.38, and the latest Kaggle submission scored 0.41, according to Cointribune.
Did the Neuraxon experiments require GPUs?
No. Cointribune says the experiments used about 500 CPU cores, with no GPU and no pretrained model.
How large was the published dataset?
The dataset behind the experiments contains 603,054 rows, according to Cointribune’s summary.
Which outlets reported this story?
This is an original Token Times brief that synthesises reporting from Cointribune. It is not a copy of any one article — follow the source links above for the original reporting.
Original summary — not financial advice. This is an original Token Times brief that summarises a development reported elsewhere and adds live market data for context; it is not the original article and reproduces no part of it. Follow the source link above for full details. Crypto is volatile and high-risk — always do your own research.