Multiverse Computing released Quasar 438B, a European reasoning model

Multiverse Computing published Quasar 438B on 2 September 2026, its first large model release. Multiverse Computing describes it as a reasoning model built for enterprise-scale agents and coding, running in English and Spanish and available through the company's CompactifAI API.

The numbers Multiverse published

Multiverse Computing says Quasar scores 43 on the Artificial Analysis Intelligence Index v4.1.1, a composite of nine evaluations that includes GDPval-AA v2, τ³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience and AA-LCR. The company positions that score as the highest among European models on the same index, ahead of Mistral Medium 3.5 at 30 and NVIDIA Nemotron 3 Ultra at 38, in a field led by Claude Opus 5 at 63.

On latency, Multiverse Computing reports Quasar returns 500 tokens including reasoning time in 15.3 seconds. The company names three faster models in the same comparison: Nemotron 3.5 Lightning at 9.4 seconds with an index score of 24, Gemini 3.5 Flash-Lite at 10.8 seconds with 37, and Gemini 3.7 Flash at 11.5 seconds with 56. Only the last of those is both faster and scores higher on the composite index, according to the release.

On long-context reasoning, Multiverse Computing reports 75.0 on AA-LCR, which tests reasoning across long documents, level with Grok 4.6 (high) at 75.0 and within a point of Claude Opus 5 at 75.7. On Terminal-Bench v2.1, which puts agents inside real terminal environments, Multiverse Computing reports 69.3, leading Mistral Medium 3.5 by 18.7 points and Nemotron 3 Ultra by 15.4, and trailing Claude Opus 5 at 89.1.

Access

Quasar is available through the CompactifAI API. The release names sign-up at dashboard.compactif.ai as the way to start sending requests, and it points to Artificial Analysis for the third-party ranking numbers cited above. Multiverse Computing does not publish per-token pricing in the announcement.

The release positions Quasar for software development agents, technical copilots, research systems and workflow automation. Multiverse Computing names English and Spanish as the supported languages and points to the response-time figure as evidence Quasar can work inside an interactive product rather than a batch job.

A team choosing a reasoning model for a coding agent has had a short list of top models, most of them hosted in the United States. Quasar adds a European provider with a published Terminal-Bench score above 69, which puts it inside the range where an agent can complete meaningful multi-step terminal tasks. Whether it replaces a model a team already uses depends on the same trade-off a team makes for any model: measured accuracy on the workload that matters, versus price and latency at real concurrency.

Two caveats apply. Every benchmark cited here is Multiverse Computing's own reporting of third-party scores, and the release page shows them without independent replication. The company also says on the page that the Terminal-Bench result is "the evaluation with the most headroom for Quasar, and it is where the next round of work is aimed," which is a direct statement that its agentic terminal work is not yet at the level of the top group led by Claude Opus 5.

Source

• Introducing Quasar 438B: Europe's Leading AI Model, Multiverse Computing, 2 September 2026