model
NVIDIA's 124B-parameter original model above the 100B line in 2026.
No evidence-backed relationships are recorded.
Researchers writing on LessWrong found only Nemotron 3 Super and Qwen3 32B changed refusals when test suspicion was removed from their reasoning traces. In most other models, test talk in a trace looks like general caution, so it is weak evidence of eval gaming.
Reality
NVIDIA says multi-agent systems burn up to 15 times the tokens of a standard chat, and Nemotron 3 Super is its open-weight attempt to make each of those tokens cheaper to produce. The efficiency figures come with NVIDIA's own hardware and its own predecessor as the baselines.
Reality
Koa is a post-trained Nemotron 3 Super running inside Salesforce's trust boundary, taught by reinforcement learning on simulated enterprise workflows. The case the company makes for it rests on consistency and inspectable provenance.
Reality
The four-stage loop generates its own ground truth by backtesting each candidate detection against the red agent's captured telemetry, which means the fidelity of the cloned environment sets the ceiling on what it can find.
Reality
Hugging Face now hosts 2.96 million public model repositories. The ones anyone actually pulls number in the tens of thousands, and the monthly parameter ceiling has been set in China all year.
Reality