MLCommons

MLPerf Training v6.0 results are live! For the first time: two Mixture-of-Experts (MoE) benchmarks reflecting where the AI training frontier actually is. 📍 DeepSeek V3 — 671B params (largest in MLPerf history) 📍 GPT-OSS 20B — 21B params mlcommons.org/2026/06/mlperf… 1/4

There's an AI Reliability Map. Most of it is still empty. Benchmarking clusters in a few cells. Enterprise AI readiness needs the full grid. AIRR is mapping what others skip: mlcommons.org/2026/04/airr-m…

MLCommons is introducing an Edge Agentic Inference benchmark for MLPerf Inference v6.1. Single accelerator. One user. Multi-turn tool-calling. Hard 32K context wall. Model: Qwen3.6-27B Q4_K_M Accuracy gate: BFCL v4 Deadline: July 31, 2026 Read more: mlcommons.org/2026/07/mlperf…

MLPerf Inference now measures multi-turn agents. 990 trajectories, Kimi K2.6 + Qwen3.6-35B-A3B, Pareto-curve performance, three-level accuracy. Built on MLPerf Endpoints. mlcommons.org/2026/07/agenti… MLPerf #AgenticAI #LLM #Inference #Kimi #Qwen #MLCommons

AI is in doorbells, hearing aids, and factory sensors. But how do you fairly compare a $1 MCU against a neural accelerator? MLPerf Tiny v1.4: 9 orgs, 25 configs, all measured the same way — standardization makes progress possible. mlcommons.org/2026/07/mlperf…

Standardized benchmarking is the only way to compare AI systems fairly at scale. MLPerf Training v6.0 results are live, featuring: -11,000+ accelerator systems -New first-time submitters -Verified performance on the world's most demanding workloads 🔗 bit.ly/4faJ8mR

@nvidia | @Oracle | @QuantaQCT | @scitix | @Supermicro | @__tinygrad__ | TTA | @Vultr Welcome, first-timers: Inventec, Netweb Technologies, TTA, and Vultr! 🎊 4/4

Huge congrats to all MLPerf Training v6.0 submitters! @AMD | @ASUS_OFFICIAL | @Cisco | @Dell | @Fujitsu_Global | @GigaComputing | @HPE | Inventec | KRAI | @LambdaAPI | @Lenovo | Mangoboost | @Azure | @MiTACcomputing | @nebiusai | @netwebtech | 3/4

MLPerf Training v6.0 results are live! For the first time: two Mixture-of-Experts (MoE) benchmarks reflecting where the AI training frontier actually is. 📍 DeepSeek V3 — 671B params (largest in MLPerf history) 📍 GPT-OSS 20B — 21B params mlcommons.org/2026/06/mlperf… 1/4

Tonight at #VLSI2026: MLCommons' David Kanter joins the Evening Panel "AI: Grand Vision or Grand Delusion?" alongside panelists from AMD, SK Hynix, Rapidus & Oxmiq Labs. 8–10 PM, Tapa 1-3.

MLPerf Mobile v6.0 introduces new generative AI benchmarks for running LLMs (Llama 3.1 & 3.2, including the new 1B and 3B models) natively on mobile devices. Test your on-device inference performance. Available on GitHub, iOS & Android: bit.ly/43dlMGE

The security world's "find it → patch it → disclose it" model doesn't work for AI. You can't patch a released open-weight model. The weights are already out there — forever. MLCommons is building the disclosure standard AI evaluation actually needs. bit.ly/43t8R3t

AI systems co-design is too fragmented. Enter MLCommons Chakra (#MLSys2026): an open execution trace ecosystem to bridge software & hardware without exposing IP. Native in @PyTorch, NVIDIA NeMo, & vLLM. bit.ly/4vkYZEP

Meet GeoCroissant. Built on MLCommons Croissant, it adds Earth observation-specific metadata—from coordinate systems to spatial resolution—to give you better traceability and more reproducible workflows for agentic AI pipelines. bit.ly/3PTLywz

Introducing the 2026 @MLCommons Rising Stars! 🌟 We’ve selected 39 outstanding early-career researchers from 26 global institutions who are shaping the future of ML systems, hardware-software co-design, and trustworthy AI. Meet the cohort: bit.ly/3Ru3ONl #AI #MLCommons

The median AI benchmark longevity score is 5/100. AILuminate scored 75—but even that degrades over time. To fix this, the @MLCommons AIRR team built the Continuous Prompt Stewardship System to keep risk evaluation fresh and reliable. bit.ly/3On4jrz

What does AI reliability actually require? It comes down to consistently following the right behavioral rules—even under adversarial attack. Meet the AI Reliability Map to guide pre-deployment testing. Explore the framework: bit.ly/4mG7erO #AIReliability #AI

Do tools like OpenClaw signal a turning point for mainstream AI adoption? MLCommons' Dave Graham debated that and more on the Utilizing AI podcast. What do you think? bit.ly/4uJj4Va #AgenticAI #AI

MLPerf Training v6.0 has added GPT-OSS 20B. With 21B total parameters (but only 3.6B active per token), this new sparse MoE pretraining benchmark is designed specifically for accessibility—it can run on a single 8-GPU node. bit.ly/4noRr14

Mixture-of-Experts (MoE) architectures like DeepSeek-V3 are the new standard for scaling frontier LLMs. Now, that architecture is part of MLPerf Training v6.0. bit.ly/3QSteEj