Agents
MLPerf Inference v6.1: pioneering agent, VLM benchmarks
MLPerf Inference v6.1 introduced the first agentic workload run on datacenter hardware and evaluated a model with more than one trillion parameters (Kimi K2), setting a new benchmark for scale. The te
MLPerf Inference v6.1 introduced the first agentic workload run on datacenter hardware and evaluated a model with more than one trillion parameters (Kimi K2), setting a new benchmark for scale. The test suite saw an 8.85 % throughput increase over v6.0, with four NVIDIA Blackwell‑Ultra GPUs delivering record Offline performance for GPT‑OSS 120B, largely thanks to software stack optimizations. In the Open division, an NVIDIA HGX B200 system ran Kimi K2 in place of Qwen3 VL, achieving a 28.5 % faster Offline rate than the best v6.0 submission while demonstrating large‑scale agentic inference on datacenter GPUs.
Related
- ThunderAgent: 2x Faster Agentic Inference for Synthetic Data Generation at Scale
- Introducing Kimi K2.6 from @Kimi_Moonshot, a multimodal agentic model with Agent Swarm scaling to 300 sub-agents and long-horizon coding sta…
- We gave a 2 hr deepdive on how to build inference engines that handle trillion token agentic workloads at @aiDotEngineer. Will drop slides a…
Source: Lambda Labs | 2026-09-16