Quasar 438B Claims Highest AA Intelligence Index of Any European Model at 43
Multiverse Computing has shipped Quasar 438B, a 438-billion-parameter reasoning model built for enterprise agents and coding. It scores 43 on the Artificial Analysis Intelligence Index v4.1.1, the highest recorded for any European-origin AI system, and is available now through the CompactifAI API.
The launch puts distance between Quasar and the two closest European competitors: Mistral Medium 3.5 at 30 and NVIDIA Nemotron 3 Ultra at 38 on the same index. Against US and Chinese frontier models, Quasar sits behind the top tier — GPT-5.6 Sol at 57 and Fable 5 at 64 — but ahead of several deployed midrange models and within reach of models commonly used in enterprise pipelines.
Key Numbers
| Benchmark | Quasar 438B | Mistral Medium 3.5 | Nemotron 3 Ultra |
|---|---|---|---|
| AA Intelligence Index | 43 | 30 | 38 |
| Terminal-Bench 2.1 | 69.3 | ~50.6 | ~53.9 |
| AA Long Context Reasoning | 75.0 | 65.3 | 71.0 |
| 500-token response time | 15.3s | 18.8s | — |
| Output speed | 176.2 t/s | — | — |
| Input / Output (per 1M) | $0.60 / $1.80 | — | — |
The 176.2 tokens-per-second throughput sits well above the 105.2 t/s median for reasoning models in the same tier, which matters for interactive agentic loops where latency compounds across tool calls.
The AA-LCR score of 75.0 matches Grok 4.6 on long-context reasoning, a benchmark that measures ability to extract and reason over information in extended documents. That parity with a top-five global model is the data point Multiverse Computing leads with for document-intensive enterprise use cases — legal analysis, financial research, multi-step code refactoring across large repos.
European Sovereign AI Context
Quasar is Multiverse Computing’s first large model. The company has historically focused on compressed AI and quantum-hybrid computing; Quasar represents a step into the 400B-plus parameter class that the major labs have dominated. The model supports English and Spanish, targeting European and Latin American enterprise deployments.
“European enterprises need models that can work through complex tasks, use tools and handle long documents, and they also need greater choice and access to powerful AI developed here in Europe,” said Enrique Lizaso, co-founder and CEO of Multiverse Computing. “Quasar brings those two requirements together.”
The CompactifAI API removes the infrastructure hurdle — organizations can integrate the model without provisioning bare-metal GPU clusters, which has historically been the forcing function pushing European enterprises toward US-hosted APIs.
Where It Falls Short
The AA Intelligence Index score of 43 still trails the frontier by 20+ points. Quasar’s Terminal-Bench 2.1 score of 69.3 is solid for a European model but below GPT-5.5 (84.7%) and Claude Opus 4.7 (80.2%) on the same harness. For organizations with frontier SWE tasks or the most demanding agentic pipelines, Quasar is not a drop-in replacement for top-tier US models.
The use case is more specific: enterprises that need European data residency, Spanish-language support, or regulatory compliance reasons to avoid US cloud providers. For those teams, a model at AA-43 with low latency and competitive pricing is a real operational choice rather than a compromise.
Multiverse Computing says further development of Quasar’s coding and agentic capabilities is planned, with additional updates to follow the launch.