Minima Blog
Research notes and systems writing on efficient inference, agent infrastructure and the economics of useful intelligence.
Article
Minima and AWS compressed Arcee’s 26B Trinity-Mini for Trainium2: a 3.3× smaller checkpoint, a 3.5× smaller KV cache, twice the replicas per trn2.48xlarge and up to 2.6× the tokens per instance.
Read articleArticle
One NVIDIA DGX Spark kept Qwen 3.6 27B and Gemma 4 31B resident behind a shared Minima endpoint. Two teams received separate Kubernetes clusters through vCluster, and a Job in each cluster called its assigned model.
Read articleArticle
Hybrid search, payload filters, and late-interaction reranking in Qdrant plus Minima-optimized inference delivered 2.92x more successful agentic RAG tasks per GPU-hour without reducing grounded quality.
Read articleArticle
Minima solved the efficiency problem inside the inference worker: 53% more tokens from Qwen3.6-27B on Blackwell. Rafay Token Factory solved the operability problem around it: a governed, token-metered service.
Read articleBenchmark with us
Send us the model, hardware and workload you care about. We will propose a benchmark with explicit quality, throughput, memory and cost gates.
Benchmark your model