Multiverse Computing launches Pulsar 16B, an open “reasoning” model built with NVIDIA architecture

Multiverse Computing said on June 22 it launched Pulsar 16B, an open reasoning model intended to help systems produce more deliberate, step-by-step outputs for a range of applications. The announcement links the model’s development to work done in collaboration with NVIDIA, and it says Pulsar 16B is powered by NVIDIA architecture.

The company described Pulsar 16B as a “reasoning” model. In general terms, that refers to language models designed to generate answers through multi-step inference rather than producing a single, direct response. Multiverse did not provide additional technical benchmarks or deployment details in the material available for review, but it framed the release as a way to broaden access to reasoning-focused model capabilities.

The specific naming in the launch points to model scale, with “16B” typically indicating a parameter count on the order of 16 billion. Larger parameter counts can translate into different performance characteristics across tasks, though the announcement information reviewed here does not include accuracy, latency, cost, or evaluation results.

NVIDIA’s role, as described in the launch coverage, centers on its architecture. For NVIDIA, these partnerships matter because demand for AI training and inference has increasingly shifted toward hardware that can accelerate both large-scale model development and day-to-day model usage in data centers and developer ecosystems.

The release of an open model is also notable for the AI sector’s broader pattern. Open weights or open reasoning models can speed experimentation by researchers and developers, and they can create demand for compatible compute platforms. Still, the available information does not specify the licensing terms Multiverse is using, nor does it clarify whether the model is distributed as open weights, through an interface, or via another mechanism.

From an NVIDIA business perspective, the announcement fits into its strategy of expanding the “full stack” for AI, spanning GPUs and software tooling to model training and inference. By enabling third parties to launch models that reference NVIDIA architecture, NVIDIA can benefit indirectly through higher utilization of its platforms, even when NVIDIA is not the direct distributor of the model itself.

What remains unclear is how Pulsar 16B will be distributed and supported over time. The information reviewed here does not detail rollout plans, model size variants, fine-tuning availability, or how developers are expected to run it at scale, nor does it include any disclosed performance targets or customer use cases.

Investors and industry watchers will likely look next for follow-up technical documentation from Multiverse, including evaluation results and any integration guidance that ties Pulsar 16B to specific NVIDIA software or deployment paths. NVIDIA’s own product messaging around inference and developer tooling could also indicate how broadly it expects partners to adopt the architecture referenced in the launch.