Multiverse Computing's first large model — a 438B-parameter English/Spanish reasoning model for enterprise agents and coding, and Europe's top scorer on the Artificial Analysis Intelligence Index.
Last verified · 2026-09-02 · by Moe Ameen
Quasar 438B is a large language model from Multiverse Computing, a San Sebastián, Spain company known for its CompactifAI model-efficiency platform. Launched September 2, 2026, it is the company's first large model — a 438-billion-parameter reasoning model built for enterprise-scale agents, coding, and multi-step reasoning. It operates in English and Spanish and is delivered through the CompactifAI API rather than a consumer chat app.
Its calling card is a score of 43 on the Artificial Analysis Intelligence Index v4.1.1, which independent evaluator Artificial Analysis rated the highest of any European model. That puts it ahead of Mistral Medium 3.5 (30) and NVIDIA Nemotron 3 Ultra (38), and behind frontier US models such as Claude Opus 5 (63). It also posts 69.3 on Terminal-Bench v2.1 (agentic terminal work) and 75.0 on AA-LCR (long-context reasoning), and returns 500 output tokens in about 15.3 seconds — fast for a model of its size, which matters when it sits inside an interactive product.
The framing is European sovereign AI: a domestically built, high-scoring model that EU enterprises can adopt without depending on US or Chinese providers. You reach it via the CompactifAI API (sign-up at dashboard.compactif.ai), and Multiverse points organizations to a sales contact; per-token pricing was not published at launch.
For content work, keep its role straight. Quasar is a reasoning and text model — it drafts, reasons, codes, and works through long inputs. It generates no images, no video, and no audio, holds no brand voice across a set of pieces, and publishes nowhere. It produces words, and stops at words.
The cleanest way to picture the pairing: Quasar is a brain that thinks and writes; Kompozy is the body that shows up on every platform. Quasar's real strength is reasoning and drafting — and, unusually at this tier, doing it natively in both English and Spanish. Its equally real limit is that it produces text and only text. It draws no image, films no talking head, lays out no carousel, and pushes nothing to a feed. Every one of those is a first-class Kompozy output.
Here is the concrete handoff. Take a Quasar draft — a script, a hook, a full article outline — and drop it into Kompozy, where that single input fans into 18 formats Quasar cannot touch: [Persona Shorts](/glossary/persona-shorts) and HeyGen avatar video that put the script on camera, [Clipped Shorts](/glossary/content-repurposing) captioned and reframed 9:16 / 1:1 / 16:9, brand-exact [Carousels](/glossary/hyperframes) through HyperFrames, face-locked persona images via Gemini, quote graphics, a blog article, and a matching newsletter. A [Persona Brief](/glossary/persona-brief) and banned-word filters hold one voice across the whole batch, so it reads as one creator rather than a stack of one-off generations. Because Quasar is bilingual, you can run parallel English and Spanish personas from the same idea. Then [Autopilot](/glossary/autopilot) schedules and publishes the set across the eight social platforms plus blog and email behind a per-post review. Quasar writes the thinking; Kompozy makes it visible, on-brand, and live. (Note that Kompozy generates its own copy with Claude and OpenAI and its visuals with gpt-image, Gemini, and HeyGen — Quasar is an upstream drafting option you feed in as source text, not a model that runs inside Kompozy.)
Quasar 438B is Multiverse Computing's first large model, launched September 2, 2026 — a 438-billion-parameter reasoning model for enterprise agents, coding, and long-context reasoning, running in English and Spanish and delivered via the CompactifAI API. Artificial Analysis rated it the highest-scoring European model on its Intelligence Index (43).
Text and reasoning: drafted copy in English or Spanish, analysis over long documents, code, and multi-step agent tasks. It does not generate images, video, or audio, and it has no publishing. It produces words that another tool then turns into finished content.
Not on its own — it is an API-only text model with no visuals and no publishing. To turn a Quasar draft into captioned video, carousels, images, blogs, and newsletters and publish them across platforms, pair it with a content engine like Kompozy, which handles the formats, brand voice, and distribution.
Through Multiverse Computing's CompactifAI API, with sign-up at dashboard.compactif.ai and a sales contact for organizations. Per-token pricing was not published at launch, so confirm current access and cost on Multiverse's official pages.
English and Spanish. The native bilingual coverage is a real advantage for work in those languages, but there is no support for other languages at launch.