Multiverse Computing
Maker of the CompactifAI compressed models, Quasar and Carina.
Multiverse Computing, based in San Sebastian, Spain, builds compressed versions of large language models with its CompactifAI method, which uses quantum-inspired tensor networks to remove redundant weights. Its Quasar models are compressed from Z.ai's GLM family and Carina from OpenAI's gpt-oss architecture.
multiversecomputing.comCall Multiverse Computing models through Opper with one API key. Sign up without a credit card.
Models from Multiverse Computing
| Model | Context | Provider | Region | Zero data retention | Training | Input | Output | Cache read |
|---|---|---|---|---|---|---|---|---|
| 203K | EU | No | $0.11 | $0.33 | $0.0099 | |||
| 262K | US | No | $0.11 | $0.33 | $0.0099 | |||
| 203K | EU | No | $0.11 | $0.33 | $0.0099 | |||
| 131K | EU | No | $0.02 | $0.09 | Input price | |||
| 131K | US | No | $0.02 | $0.09 | Input price |
Release timeline
Every dated Multiverse Computing release, newest first — 3 of 3 run on the Opper gateway today. Releases without a published date aren't listed. See the AI model release tracker.
3 releases in the last 12 months42 days median gap35–49 days range