SLM Ensemble
SLM Ensemble is the public product published under the MegaCpp brand by Datasunrise OÜ. The concrete public model lane is NAM56R: a Mamba 3 plus Transformer hybrid with 16 routed experts, top-4 activation, about 4.73B total parameters, and about 3.03B active parameters, while hardware-specific claims stay in technical articles where they can be scoped properly.
What this page is trying to say, precisely
Datasunrise OÜ is not presenting MegaCpp as a public suite of orchestrators, debug models, shell models, and routing products. Those are implementation and research surfaces. The public commitment is simpler: one product for C++ engineering, supported by architecture notes, public samples, and a company story that is easy to verify.
That means this page avoids unsupported benchmark tables, unpublished dataset counts, and hardware promises that are not backed by public receipts here. Where deeper claims matter, they belong in dedicated technical articles with references.
The useful summary is that SLM Ensemble is the public interface to our NAM56R-centered C++ model work, while the blog, architecture docs, and public samples repository carry the narrower engineering details.
Public principles
Six things we can say cleanly in public
These are the durable themes behind the product page. They describe direction and scope without inventing a benchmark leaderboard.
C++ code work
The public product is designed around C++ authoring, review, and patch-oriented workflows rather than general chat.
Build-aware context
We care about compiler-aware context, cross-file structure, and codebase grounding because that is what makes C++ tasks real.
Hybrid architecture
Hybrid design is part of the MegaCpp research direction. We describe it publicly as an active engineering choice, not as a universal victory claim.
Evaluation discipline
Public evaluation language is kept narrow. We prefer methodology, receipts, and references over headline benchmark claims without context.
Deployment reality
Training and inference notes are published as technical articles where assumptions, hardware, and tradeoffs can be stated explicitly.
C++ specificity
Templates, build systems, refactors, review loops, and codebase navigation are treated as one product surface rather than separate public products.
Follow the deeper technical material
The product page should stay narrow. Use the blog for architecture, training, CUDA or NVIDIA-specific notes, and the documentation pages for reference material.