Multiverse Computing · 2026-09-02 · major
Quasar 438B — Multiverse Computing's first large model, built in Europe
Quasar 438B is Multiverse Computing's first large model, a 438-billion-parameter reasoning model for enterprise agents and coding. It scores 43 on the Artificial Analysis Intelligence Index, the top result for a European model.
Multiverse Computing's first large model: 438B parameters, aimed at enterprise agents and coding.
Quick facts
| Maker | Multiverse Computing |
|---|---|
| Parameters | 438 billion |
| Languages | English and Spanish |
| Speed | 500 output tokens in 15.3s |
| Agentic coding | 69.3 on Terminal-Bench v2.1 |
| Long-context reasoning | 75.0 on AA-LCR |
| Availability | CompactifAI API, access on request |
Benchmarks
| Claude Opus 5 | 63 pts | |
|---|---|---|
| Quasar 438B | 43 pts | |
| Inkling | 42 pts | |
| NVIDIA Nemotron 3 Ultra | 38 pts | |
| Mistral Medium 3.5 | 30 pts |
What is it?
Quasar 438B is the first large model Multiverse Computing has released, a 438-billion-parameter reasoning model the company built for enterprise agents and coding. It supports English and Spanish. Multiverse says Quasar is the strongest-performing European model in Artificial Analysis's latest comparison, where it scores 43 on the Intelligence Index v4.1.1.
How does it work?
The model thinks before it answers, and Multiverse's headline speed figure covers that step: 500 output tokens in 15.3 seconds, against 18.8 seconds for Mistral Medium 3.5 on the same measure. Quasar 438B is served through CompactifAI, the Multiverse API built around the company's tensor-network model compression work. The announcement does not describe the architecture or offer a weights download.
Why does it matter?
European teams that want a capable model from an EU vendor have had few choices, and Quasar 438B gives them one that Artificial Analysis ranks above Mistral Medium 3.5 and NVIDIA Nemotron 3 Ultra. Access runs through the CompactifAI API on request, so the launch is aimed at enterprises rather than individual developers. Quasar still sits well below Claude Opus 5, which scores 63 on the same index.
Who is it for?
European enterprises building agents
Frequently asked questions
- How much does Quasar 438B cost?
- Multiverse Computing has not published prices for Quasar 438B. Both the launch post and the press release describe availability through the CompactifAI API with access granted on request, and point developers at business@multiversecomputing.com rather than a public price page. Teams comparing token costs will need to ask Multiverse directly.
- Is Quasar 438B open source?
- Multiverse Computing has not released Quasar 438B weights. The announcement covers API access through CompactifAI only, and mentions no licence, no download and no self-hosting option. Anyone who needs weights they can run in their own data centre should treat Quasar as a closed, hosted model for now.
- How does Quasar 438B compare with Mistral Medium 3.5?
- Quasar 438B scores 43 on the Artificial Analysis Intelligence Index v4.1.1 against 30 for Mistral Medium 3.5, and it answers faster: 15.3 seconds for 500 output tokens including thinking time, versus 18.8 seconds. Multiverse also compares Quasar with NVIDIA Nemotron 3 Ultra, which scores 38 on the same index.
- What languages does Quasar 438B support?
- Quasar 438B supports English and Spanish, according to Multiverse Computing's announcement. No other languages are listed. Teams that need French, German or Italian coverage should check with Multiverse before committing, because the launch material makes no claim about them either way.
- What is CompactifAI?
- CompactifAI is the Multiverse Computing product that serves Quasar 438B. Multiverse describes it as an AI model compressor that uses tensor networks to make AI systems faster, cheaper and more energy efficient, offered as an inference API for both original and compressed models. The CompactifAI page carries a demo request form for teams evaluating it.
Try it
Request CompactifAI API access: business@multiversecomputing.com