Cloud storage buyers rarely get vendor-provided performance data that includes the vendor’s own weak spots. Backblaze’s Q1 2026 Performance Stats report, attempts to do exactly that, sharing benchmark results for Backblaze B2, AWS S3, Cloudflare R2, and Wasabi Object Storage across US-East and EU-Central regions, and including results where Backblaze’s own rate limits affected the numbers.
The report is the second in a quarterly series. Backblaze publishes its full methodology and states that tests run from a neutral Vultr-hosted Ubuntu virtual machine, routed through Catchpoint’s network, to avoid identifying the test account as Backblaze-owned to competing providers.
“What you’re seeing in this quarter’s data are early signals about provider strengths and weaknesses, which start to make sense once you understand the architectural decisions behind them,” said Dan Spraggins, SVP of Engineering, Backblaze. “As this dataset grows quarter over quarter, those signals will become patterns, and those patterns will fundamentally change how the industry thinks about cloud storage strategy.”
What was tested
Tests covered average upload and download times for 256KiB, 2MiB, and 5MiB files, time-to-first-byte (TTFB) for downloads, and five-minute sustained throughput tests in both single-threaded and multi-threaded configurations. Throughput tests used file sizes of 256KiB, 5MiB, 50MiB, and 100MiB. Multi-threaded tests ran 20 concurrent threads. EU-Central data originates from an Amsterdam data center.
US-East upload and download averages
Across US-East upload averages, every provider recorded lower times compared to Q4 2025. Backblaze led for 256KiB and 5MiB uploads; Wasabi led for 2MiB uploads. In US-East download averages, AWS S3 led in TTFB, 256KiB, and 5MiB categories. Backblaze took the 2MiB download category, a position AWS held last quarter.
Rate limits in the throughput data
Backblaze hit its own bandwidth caps during multi-threaded throughput tests at larger file sizes in US-East. The bandwidth cap did not affect the 256KiB or 5MiB file sizes in the multi-threaded upload test, only the larger ones.
To address the rate-limit behavior, Backblaze ran an auto-retry script and raised the bandwidth cap for affected test sizes only, keeping all other account parameters at standard tier defaults. Backblaze had an advantage in detecting its own rate limits, and that other providers may apply similar or different constraints without those being visible in the test data.
In US-East multi-threaded upload throughput, Wasabi led across file sizes. In US-East multi-threaded download throughput, Backblaze led for 256KiB, AWS led for 5MiB and 50MiB, and Cloudflare led for 100MiB.
EU-Central findings
The EU-Central region, added this quarter, produced different provider rankings than US-East. In EU upload averages, Cloudflare R2 led the 256KiB category and Backblaze led for 2MiB and 5MiB. In EU download averages and TTFB, Cloudflare led in TTFB, 256KiB, and 2MiB categories; AWS led for 5MiB.
In EU multi-threaded upload throughput, Wasabi led for 256KiB and 100MiB; AWS led for 5MiB and 50MiB. Backblaze’s EU download average times drew internal attention. Backblaze identified an issue it believes contributed to those numbers and plans to report on the impact of fixes in a future quarter.
Variance and dataset maturity
A recurring observation across both regions is the wide spread between highest and lowest throughput values within each file size category. This pattern appeared last quarter as well, and flags it as something worth tracking over time. It also cautions that the dataset is still young and that significance cannot yet be attributed to quarter-over-quarter changes in throughput results, given that the methodology changed between Q4 2025 and Q1 2026. Upload and download averages were unaffected by the methodology change and are considered comparable.
A separate pattern visible in both regions is that multi-threaded download throughput rises sharply from 256KiB to 5MiB file sizes and then flattens out for larger files.
Methodology limitations
The report lists several constraints, including a 10MiB file size ceiling on average upload and download tests due to timeout thresholds, a static origination point in NY/NJ for US-East tests, and the possibility of intermediate caching in repeated download tests. It also notes that Wasabi’s inbound connection rules affected test behavior in a prior quarter.
The full methodology, raw data tables, and per-file-size clustering charts are included in the report.
