Largest MLPerf Training v5.0 submission
Company: CoreWeave
The claim, verbatim
CoreWeave delivered the largest-ever MLPerf Training v5.0 submission on GB200 instances, 34x larger than other submissions and 4.5x more performant than the best GB200 submission from any other enterprise.
Source (primary)
coreweave 8-K exhibit (coreweave2q25earningspress.htm) (SEC EDGAR (8-K Ex), press_release)
View cached copy (2026-10-07)Live source ↗
Quote: “Delivered the largest-ever MLPerf Training v5.0 submission utilizing NVIDIA Blackwell GB200 instances. The breakthrough submission was 34X larger than other submissions and 4.5x more performant than the best GB200 submission from any other enterprise”
How we checked this
Checked on October 9, 2026. The cited source supports every part of this claim.
The release makes each of these statements about CoreWeave's MLPerf Training v5.0 submission.
Confirmed in the source:
- CoreWeave delivered the largest-ever MLPerf Training v5.0 submission
- The submission used NVIDIA Blackwell GB200 instances
- The submission was 34x larger than other submissions
- The submission was 4.5x more performant than the best GB200 submission from any other enterprise
What we did: Read our cached copy of SEC EDGAR (8-K Ex) (https://www.sec.gov/Archives/edgar/data/1769628/000176962825000039/coreweave2q25earningspress.htm) in full (24,727 characters, retrieved October 7, 2026) and checked each assertion in the claim against it.
Additional evidence
confirms coreweave 8-K exhibit (coreweave2q25earningspress.htm)
Quote: “Delivered the largest-ever MLPerf Training v5.0 submission utilizing NVIDIA Blackwell GB200 instances. The breakthrough submission was 34X larger than other submissions and 4.5x more performant than the best GB200 submission from any other enterprise”