Saturn Cloud has partnered with Tileward to bring compressed open models to its token factory. Operators running the token factory get the same models in a fraction of the GPU memory, with comparable accuracy on Tileward’s evals, so each GPU serves more tokens.
Sell inference under your own brand
Saturn Cloud’s token factory turns GPU fleets into managed inference and fine-tuning services that operators sell to their own customers, under their own brand. Tenant isolation, quotas, metering, and enterprise governance are built in. It runs on NVIDIA Dynamo, in public cloud, private cloud, or on-premises.
On Kimi K3, with recorded coding-agent traffic across 16 live agent sessions on NVIDIA GB300 GPUs, Saturn Cloud served 7,380 total tokens per second per GPU, against 7,281 for the SemiAnalysis InferenceX benchmark recipe on the same GPUs.
Saturn Cloud also helps operators optimize input, cached, and output token rates across every model they serve, including custom models.
Free up GPU memory
Tileward, a member of the NVIDIA Inception program for startups, offers compressed models that are up to 78% smaller than the published weights. Tileward tests accuracy across benchmarks such as MMLU, SWE-bench Verified, and AgentBench, as well as across different workload types.
Operators serve Tileward’s compressed weights in place of the published weights, and their serving stack, tenants, rate cards, and brand stay the same. The freed memory lets them run more copies of each model, fit large models on fewer GPUs, and serve more concurrent users with longer contexts. Serving more tokens from the same GPU lowers the cost of each token, which gives operators more margin on input, cached, and output rates, or room to price lower.
Available now
Tileward’s compressed versions of gpt-oss-120b, GLM-5.3, and DeepSeek-V4-Flash are available now on Saturn Cloud’s token factory. Neoclouds, AI factory operators, telcos, sovereign AI providers, and enterprises running their own GPUs can use them to serve more tokens from the GPUs they already run. To add them to a deployment, talk to us.



