Overview
- Multiverse announced Quasar 438B on Wednesday, Sept. 2, 2026, releasing the model through its CompactifAI API as a 438‑billion‑parameter system built for coding, multi‑step agents, and long‑document workflows.
- Public benchmarks show Quasar scored 43 on the Artificial Analysis Intelligence Index and 69.3 on Terminal‑Bench v2.1, and measured end‑to‑end output speed at roughly 183 tokens per second or about 15.3 seconds for a 500‑token response.
- Multiverse credits its CompactifAI compression for Quasar’s efficiency and says the technique can shrink models by 80–95 percent, but the company has not disclosed how much Quasar was compressed or which base model it started from.
- Quasar offers a 1,000,000‑token context window and supports English and Spanish, but independent testing is limited because the model is proprietary and available only via Multiverse’s API, preventing weight inspection or on‑prem deployment.
- The release follows Multiverse’s large Series C funding and positions the company in Europe’s drive for sovereign AI, yet real‑world agent latency, repeated‑call costs and hardware requirements remain open questions that will shape enterprise adoption.