Quasar 438B Tops Global AI Benchmark Index

Multiverse Computing’s Quasar 438B became the first European-built model to top the Artificial Analysis Intelligence Index on Sept. 2, edging out all regional rivals on a closely watched global AI benchmark.

Key Takeaways

Quasar 438B topped the Artificial Analysis Intelligence Index on Sept. 2, becoming the first European-built model to do so
Multiverse Computing announced a collaboration with Qualcomm aimed at running efficient AI models inside data centers
At 438 billion parameters, the model’s size sits close to Meta’s Llama 3.1 405B
Multiverse has not published pricing, availability, or a technical paper detailing how it trained or compressed the system

Quasar 438B Climbs Past Rivals On A Global AI Index

The Spanish startup announced the result alongside a Qualcomm (QCOM) collaboration aimed at running efficient AI models inside data centers. Neither side disclosed financial terms.

The index pools results from independent coding, math and reasoning tests into one composite score, draws from a rotating batch of public benchmarks, and refreshes as labs ship new checkpoints, so the model’s rank could shift once rivals publish updates.

The Quantum Trick Behind Multiverse’s Compression Pitch

Multiverse built its business on tensor networks, a mathematical structure first developed to simulate quantum systems. The technique strips redundant parameters from a trained network without full retraining, cutting compute costs.

At 438 billion parameters, the model’s headline size sits close to Meta‘s Llama 3.1 405B. Multiverse’s pitch rests on compressing such systems rather than matching their scale outright.

Also Read: Quasar 438B Beats Every European AI Model With a 43-Point Score

Europe’s Long Chase For A Homegrown Frontier Model

The result lands amid a broader European Union push to build AI on the continent, reducing reliance on OpenAI, Google, or Chinese labs.

A European model that ranks competitively on an international index gives regulators and buyers in the bloc a fallback option. Sept. 2’s benchmark placement is the clearest public proof point for Multiverse’s compression pitch so far.

What Multiverse Still Has Not Said

Multiverse has not published pricing, availability, or a technical paper detailing how it trained or compressed the system.

Artificial Analysis, the index’s operator, had not issued independent commentary on the ranking as of Sept. 3.

The Qualcomm collaboration remains similarly undefined, with no timeline for when compressed models might run inside live data centers. Whether Quasar 438B keeps its rank depends entirely on how fast US and Chinese labs answer with new checkpoints.