Magnum Enterprise
Private, single-tenant inference for industrial engineering. Magnum 1.5 runs auditable open-weight models inside mutexer’s execution harness, giving Rivets multimodal schematic and code reasoning without sending plant data to external model providers.
120tok/s
Generation speed in one conversation
550tok/s
Total generation across 16 conversations at once
2.0s
Warm reply on a 100K-token conversation
256K
Token context window per conversation
Fig. 1Our own measurements on a single reference GPU, November 2026, not vendor headline figures. A warm conversation answers from cache instead of re-reading its whole history.
Magnum 1.5 is provisioned in your own private cloud. Source code, prompts, and model output stay inside your environment. There is no third-party cloud model in the loop, so there is nothing to leak, log, or subpoena elsewhere.
Magnum Enterprise buys a pool of concurrent sessions, not a credit meter. Your engineers can generate as much code as they need without a per-token bill that scales with adoption. Budgeting stops being a guess.
Magnum 1.5 is built on open-weight foundation models. The weights can be inspected and the deployment can be validated against your own security and compliance requirements. No closed API, no vendor black box in your supply chain.
Benchmark results on two standard agentic-coding tests.
| Benchmark | Magnum 1.5 | Claude Sonnet 5 |
|---|---|---|
| SWE-bench Proreal-world issue resolution | 61.7% | 63.2% |
| Terminal-Bench 2.1agentic terminal use | 73.0% | 80.4% |
Magnum 1.5 scores are the published results of its open-weight base model. Claude Sonnet 5 scores are from Anthropic's Claude Sonnet 5 System Card (June 2026). Each provider ran its own evaluation with a different agent harness.
Magnum 1.5 is available with the Magnum Enterprise tier. Talk to us about provisioning a dedicated, private cloud inference engine for your team.