Quasar 438B is Europe’s new heavyweight: a 438-billion-parameter model from Spanish company Multiverse Computing that just became the highest-scoring European model Artificial Analysis has ever evaluated. The part that matters: it wasn’t built to be a chatbot — it was built for AI agents: multi-step workflows, long-document reasoning, tool calling. Here’s the honest read on the benchmarks, the speed story most people miss, and what a million tokens of context is actually for.
Short answer
- Built by Multiverse Computing (Spain): 438B parameters, designed for agents — workflows, long documents, tool calling.
- Artificial Analysis Intelligence Index: 43 — their highest-scoring European model to date (Claude Opus 5 leads overall at 63).
- Speed: a 500-token response including reasoning in 15.3 seconds — only three models tested faster, and just one was faster AND smarter.
- Context: 1 million tokens (~1,500 pages), scoring 75.0 on long-context reasoning — level with Grok 4.6 High, about a point under Opus 5.
What Quasar 438B is (and who built it)
Multiverse Computing is a Spanish AI company, and Quasar 438B is its first major release — a 438-billion-parameter model aimed squarely at the agent era rather than the chat era. That framing matters when you read the numbers: the design targets are multi-step workflows, reasoning across huge documents, and reliable tool calling — the jobs agents actually do — not winning trivia contests.
On the Artificial Analysis Intelligence Index — a combined benchmark across nine areas including agentic tasks, coding, scientific reasoning and long-context understanding — Quasar scored 43, making it the highest-scoring European model they’ve evaluated. For placement: Mistral Medium 3.5 scored 30, NVIDIA Nemotron 3 Ultra 38, Inkling 42 — and Claude Opus 5 still leads the field at 63. So no, this isn’t suddenly the smartest model in the world. For a first major release, and for Europe, it’s a genuinely strong debut — and that’s the honest framing.
Quasar 438B benchmarks: the speed story most people miss
Raw intelligence isn’t everything — for agents, speed compounds. Quasar produces a 500-token response, reasoning time included, in 15.3 seconds; only three models in the comparison were faster, and only one of those was also smarter. Why that matters: an agent task isn’t one model call. It reads your request, plans, calls a tool, reads the result, decides again, calls another tool, checks, adjusts, finishes — ten to twenty separate calls. Multiply a slow model through that chain and your workflow crawls; Quasar’s spot on the speed-versus-intelligence curve is exactly where agent builders should be looking.
| Model | Artificial Analysis Index |
|---|---|
| Claude Opus 5 | 63 |
| Quasar 438B | 43 — top European score |
| Inkling | 42 |
| NVIDIA Nemotron 3 Ultra | 38 |
| Mistral Medium 3.5 | 30 |
Sourcing, stated plainly: these figures are Artificial Analysis’s published benchmark results as covered in my video (early September 2026) — scores move as models update, so treat the table as a snapshot. And the launch coverage I reviewed didn’t detail pricing or access routes — check Multiverse Computing directly before building plans on it.
🔥 Want this set up without the guesswork? Long-context agent workflows like these, built for your business is exactly the kind of thing we set up together inside the AI Profit Boardroom — 3,700+ members, four live calls a week, daily tutorials, done-for-you templates and a 30-day roadmap. Prefer 1-on-1 help? Book a free SEO strategy session and we’ll map it out for your business.
Why Quasar 438B is built for agents: the 1M-token context
The context window is the feature with business consequences: 1 million tokens — roughly 1,500 pages — and a 75.0 on the long-context reasoning benchmark, which tests connecting and reasoning across that material, not just retrieving from it. That score matches Grok 4.6 High and lands about a point under Claude Opus 5. In practice, that’s feed-it-everything territory: months of coaching-call transcripts, every onboarding form and support thread, an entire knowledge base — then one prompt asking it to find the ten biggest member problems, group them by business type, and recommend the fix for each.
That’s the shape of workflow I’d run it in: a member-success agent that reads everything a community produces and writes personalised next steps per member, or a call-intelligence agent that turns each coaching call into FAQs, tutorial ideas and follow-ups — then uses tool calling to actually post and send them. Where does it sit in a stack? As a specialist: pair a long-context workhorse like this with your daily-driver models — my auxiliary models guide covers exactly that mixed-stable pattern, and the open-source line-up holds the free end.
The bottom line on Quasar 438B
Quasar 438B is the strongest European debut Artificial Analysis has measured: not an Opus-beater at 43 versus 63, but fast where agents need speed, huge where agents need memory, and honest about what it’s for. If your workloads look like long documents plus many-step tool use — the unglamorous heart of business automation — it just earned a spot on your shortlist, and Europe finally has a flagship worth benchmarking against.
FAQ: quasar 438b
What is Quasar 438B?
A 438-billion-parameter model from Spanish company Multiverse Computing, built for AI agents — multi-step workflows, long-document reasoning and tool calling.
How good are Quasar 438B’s benchmarks?
It scored 43 on the Artificial Analysis Intelligence Index — their highest European score to date; Claude Opus 5 leads overall at 63.
How fast is Quasar 438B?
A 500-token response including reasoning in 15.3 seconds — only three tested models were faster, and only one was faster and smarter.
How big is the context window?
1 million tokens — roughly 1,500 pages — scoring 75.0 on long-context reasoning, level with Grok 4.6 High and about a point under Opus 5.
Is Quasar 438B better than Claude?
On combined intelligence, no — 43 versus Opus 5’s 63. On the speed-and-context profile agents need, it’s genuinely competitive.
Who is Quasar 438B for?
Agent builders with long-context workloads — communities, support archives, document-heavy automation — ideally as a specialist inside a mixed model stack.
Next step: if you want an agent stack that picks the right model per job working for you this week, join the AI Profit Boardroom for the full walkthroughs and live help — or book a free SEO strategy session and I’ll point you at the fastest path for your situation.
About Julian Goldie: SEO agency owner with 10+ years in SEO, 394K+ subscribers on YouTube, a 100% job-success score on Upwork, 75K+ members across his communities, and author of a best-selling SEO book. He runs the AI Profit Boardroom community and offers a free SEO strategy session.
Related reading
- Hermes Agent Best Auxiliary Models: The Mixed Stable
- Best Open Source Models For Hermes Agent
- GPT-6 Astra + Hermes Agent: Setup, Cost & Verdict
Last updated September 2026. This is the living guide to quasar 438b — it gets updated as the tools change.
