Glonce

Multiverse Computing launches Quasar 438B, Europe's leading model

modelsThe Glonce editorsHacker News🔥62Glonce rating62Impact45Novelty55Relevance100Surprise40Players35Credibility80
Multiverse Computing launches Quasar 438B, Europe's leading model

Multiverse Computing has released Quasar 438B, the first large model the company has shipped and a reasoning system aimed at enterprise-scale agents and coding. It runs in English and Spanish and is available through the CompactifAI API. On the Artificial Analysis Intelligence Index v4.1.1, a composite of nine evaluations (GDPval-AA v2, tau-3-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience and AA-LCR), Quasar scores 43, which the company presents as the highest result of any European model measured. That places it 13 points ahead of Mistral Medium 3.5 (30), 5 ahead of NVIDIA Nemotron 3 Ultra (38 in this comparison, though a later paragraph in the same source gives Nemotron 3 Ultra as 36), and just ahead of Inkling (42). The field is led by Claude Opus 5 at 63, putting Quasar about 20 points behind the frontier.

On response time, Quasar returns a 500-token answer, thinking time included, in 15.3 seconds. Only three models in the comparison answer faster: Nemotron 3.5 Lightning at 9.4 seconds (index score 24), Gemini 3.5 Flash-Lite at 10.8 seconds (score 37), and Gemini 3.7 Flash at 11.5 seconds (score 56). Of those three, only Gemini 3.7 Flash also outscores Quasar on the index. Among the models that do outscore Quasar, only two answer in under 25 seconds; the rest take between 38 and 156 seconds. Set against slower rivals, Mistral Medium 3.5 scores 30 in 18.8 seconds, Nemotron 3 Ultra scores 36 in 25.7 seconds, and Inkling scores 42 but needs 48.3 seconds, more than double Quasar's time.

On AA-LCR, a long-context reasoning benchmark, Quasar scores 75.0, level with Grok 4.6 (high) at 75.0 and within a point of Claude Opus 5 (75.7) and Qwen3.8 2.4T A95B (75.3). It leads Nemotron 3 Ultra by 4.0 points and Mistral Medium 3.5 by 9.7. Multiverse Computing calls long-context handling the area where Quasar comes closest to the frontier group. On Terminal-Bench v2.1, which tests agents working in real terminal environments, Quasar scores 69.3, ahead of Mistral Medium 3.5 by 18.7 points and Nemotron 3 Ultra by 15.4, but well behind Claude Opus 5's 89.1. The company names this the benchmark with the most headroom and says its next round of work targets it.

Multiverse Computing frames Quasar as a step from its efficiency-focused compression work (the basis of the CompactifAI platform) into the 400 billion-plus parameter class, built for multi-step agent work, tool use, code execution and large context without frontier-model latency. The source gives no release date, no pricing for the API, no detail on Quasar's actual parameter count, training data or architecture beyond the '438B' name, and no information about which company builds the Inkling model used throughout the comparisons.

Key facts

  • Quasar 438B is Multiverse Computing's first large model, a reasoning system for enterprise agents and coding, available through the CompactifAI API in English and Spanish.
  • It scores 43 on the Artificial Analysis Intelligence Index v4.1.1, the highest of any European model measured, but about 20 points behind the field leader, Claude Opus 5, at 63.
  • It answers a 500-token prompt in 15.3 seconds; only three compared models are faster, and just one of those, Gemini 3.7 Flash, also scores higher on the index.
  • On AA-LCR (long-context reasoning) Quasar scores 75.0, within a point of Claude Opus 5, which the company calls its closest approach to the frontier group.
  • On Terminal-Bench v2.1 it scores 69.3, well behind Claude Opus 5's 89.1; Multiverse Computing names this the benchmark with the most room to improve.

Why it matters

Quasar 438B is Multiverse Computing's first move into large-scale reasoning models, and the company built its reputation on making AI more efficient and deployable rather than on frontier-scale training. Landing at 43 on the Artificial Analysis Intelligence Index, ahead of every other European model in the comparison, is a real claim about where the region's best current model sits: still roughly 20 points behind Claude Opus 5's 63, but ahead of Mistral Medium 3.5, NVIDIA Nemotron 3 Ultra and Inkling. The response-time numbers matter alongside the score: at 15.3 seconds for a 500-token answer, Quasar is faster than every model that beats it on the index except one, Gemini 3.7 Flash.

Who it affects

The model targets organizations running coding agents, technical copilots, research systems and workflow automation, where the Terminal-Bench and long-context results speak directly to holding context and executing multi-step tasks. English and Spanish support is aimed at European and international enterprise teams that need reasoning capability without narrowing deployment to a single language. Developers can test the model directly through the CompactifAI API rather than deploying their own infrastructure.

How to use it

Quasar 438B is reached through the CompactifAI API, with sign-up at dashboard.compactif.ai. The source gives no pricing, no context-window size and no availability details beyond the API itself, so none of that can be stated here.

How solid is it

The benchmark figures come from Multiverse Computing's own announcement, citing Artificial Analysis as the source of the underlying evaluations and charts. The article's own numbers are not fully consistent: it states NVIDIA Nemotron 3 Ultra's Intelligence Index score as 38 in the introductory comparison, then as 36 later in the same piece, without reconciling the two. That inconsistency sits inside the source Multiverse Computing published, not introduced by this retelling.

Risks and caveats

The announcement gives no release date, no pricing for the CompactifAI API, and no detail on Quasar 438B's actual parameter count, training data or architecture beyond the '438B' name and its placement in a '400 billion-plus' class. It also does not say which company or lab builds Inkling, one of the models used throughout the comparisons. The scores are self-reported in a company blog post rather than published by an independent party, and the unresolved Nemotron 3 Ultra discrepancy (38 versus 36) is a reason to treat the exact figures with some caution even though the overall ranking is unlikely to be affected.