Open, peer-reviewed measurement for today’s AI buyers: MLPerf in the EU’s AI Gigafactories tender, and Endpoints v1.0 opens for submissions
MLCommons ·
Rolling submissions, mandatory power reporting, and a place in the EU’s AI Gigafactories tender: measurement buyers can trust. The post Open, peer-reviewed measurement for today’s AI buyers: MLPerf in the EU’s AI Gigafactories tender, and Endpoints v1.0 opens for submissions appeared first on MLCommons .
As AI adoption is growing across the economy, AI buyers are asking a more complex set of questions about AI systems than simply a chip’s speed. They want to know what performance a deployed service will deliver, what it will cost for given workloads, and how much energy it will use. MLPerf has been the standard-bearer for measuring AI system performance since 2018; it’s evolving to address these more complex questions for every AI buyer.
First, out of Europe: MLPerf has been chosen as the yardstick for evaluating training and inference performance in bids for the EU’s call for AI Gigafactories . Second, MLPerf Endpoints v1.0 is open for submissions, formalizing our transition to this new era of answering complex AI purchasing questions in the open, standardized way we pioneered.
These two updates point the same way: industry, government, and buyers are converging on open, peer-reviewed measurement as the basis for AI adoption decisions.
In July, the European Union opened its call for AI Gigafactories : up to seven large-scale AI computing facilities, backed by up to €10 billion in EU and national funding and expected to unlock more than €30 billion in total investment. The tender is run as a joint procurement by the EuroHPC Joint Undertaking and national contracting authorities, with bids due November 12, 2026.
The tender specifications name MLPerf as the standard for measuring bidders’ technical performance. Proposals are scored on MLPerf Training v5.1 and Inference v5.1 benchmarks. Across the two suites, the tender names more than half a dozen models drawn from MLPerf’s standard workloads: large models such as Flux.1 and Llama on the training side, and Llama 3.1 405B, DeepSeek R1, and SDXL 1.0 for inference. The requested information builds directly on MLPerf’s standard performance metrics — time-to-train for training, time to first token, and time per output token for inference — supplemented by additional details such as accelerator utilization. Bidders must report these results both for the capacity they deliver in Phase 1 and for their planned Phase 2 expansion.
When public money on this scale is at stake, procurement teams need performance claims they can trust and compare. Open, peer-reviewed results provide exactly that.
“MLPerf is designed to help inform buyers as they navigate deploying AI capabilities. We are honored to see our work guiding the European Union as it builds toward leadership in AI and empowers its citizens and companies for the future.” – David Kanter, Head of MLPerf
MLPerf Endpoints tests AI the way it is actually consumed: through the live API that applications call to reach a model. The system under test is simply a URL that the Endpoints client talks to over standard interfaces. That means that buyers can measure hardware vendors, hyperscalers, neoclouds, and managed API providers on equal footing, whether they run bare metal, a managed cloud service, or a containerized software stack.
Instead of a single peak number, Endpoints sweeps each system across a range of loads and reports a full performance profile. Results cover concurrency, interactivity (tokens per second per user), system throughput, and time to first token measured at the 90th percentile (TTFT P90), so buyers can match a verified operating point to their own applications and service-level targets. In a first for the MLPerf family, Endpoints 1.0 introduces mandatory provisioned power measurement, demonstrating a commitment to energy efficiency. This helps buyers understand and compare energy efficiency and the cost profile for each solution. Each result appears as a pareto frontier on the LoadStair visualizer, showing where a system holds steady and where it hits a performance cliff.
Version 1.0 is now available, building on the v0.7 foundation release from July. It covers generative AI workloads including DeepSeek-R1, GPT-OSS 120B, and Llama models, plus new agentic coding workloads using DeepSeek v4 Pro, Kimi K3, and Qwen 3.6.
Endpoints now offers rolling submissions, so members can publish a peer-reviewed result as soon as they ship a new chip or software optimization rather than waiting for a fixed round. Every result still goes through MLPerf’s peer review and audit process. We expect the first v1.0 results to be published in November.
Ready to submit? MLCommons members can find the submission guide and rules at mlcommons.org/benchmarks/endpoints . Not a member yet? Get involved .
The post Open, peer-reviewed measurement for today’s AI buyers: MLPerf in the EU’s AI Gigafactories tender, and Endpoints v1.0 opens for submissions appeared first on MLCommons .