SFCompute vs Tensorlake

Sandbox Benchmark Comparison · Updated November 20, 2026

SFCompute was faster in both shared sandbox benchmarks.

Head to head

  • Time to interactive · lower is better
    SFCompute faster3.66x
    SFCompute
    621ms
    Tensorlake
    2.28s
  • Total workload time · lower is better
    SFCompute faster5.49x
    SFCompute
    39.91s
    Tensorlake
    219.25s

Burst TTI

November 20, 2026

SFCompute

100% success
Median
621ms
p95
765ms
p99
766ms

Tensorlake

100% success
Median
2.28s
p95
2.34s
p99
2.36s

Median time to interactive over time

15 runs · September 4, 2026 to November 20, 2026
SFComputeTensorlake

DAX

November 20, 2026

SFCompute

100% success
Median
39.91s

Tensorlake

100% success
Median
219.25s

Median total workload time over time

6 runs · September 16, 2026 to November 20, 2026
SFComputeTensorlake

Cost per DAX run

SFCompute

Per GiB-hour (parkable)
$0.0081per run
39.91s median at $0.7315/hr

Rollout sandboxes at $0.04572/GiB-hr while active (16 GiB); parked/idle drops to ~$0.000139/GiB-hr. Interactive cpu-8 node is $0.4824/hr with a 15-min minimum.

Source pricing page

Tensorlake

Active CPU + allocated RAM
$0.049per run
219.25s median at $0.8000/hr

$0.07/core-hr active CPU + $0.015/GB-hr RAM (on-demand rates; Pro is ~40% lower).

Source pricing page
Compare sandbox costs across all providers
Work at SFCompute or Tensorlake? See where you stand every day, with full run logs, not just on benchmark Friday.Get stealth benchmarks

Methodology

  • One workload, every provider. The same script, payload, and model hit every provider, addressed the way its own API expects. Nobody gets a tuned or special path.
  • Repeated on a schedule. Every suite runs many iterations per provider, automated on a recurring schedule, so results reflect steady state rather than one lucky request.
  • Raw results published. Every run is committed as JSON to the public benchmarks repo, next to the code that produced it. Nothing is cherry-picked or held back.
  • How this comparison is built. Each benchmark compares the two providers' medians; results within 5% count as roughly even. A comparison is published when both providers have at least one successful run in the same benchmark from the last 30 days. A benchmark only gets a verdict when both providers completed at least 50% of its runs. If a provider’s latest run of a benchmark failed, its most recent successful run from that window is used instead.
Read the full methodology

Frequently asked questions

Have another question? Email us.

SFCompute was faster in both shared sandbox benchmarks. In Burst TTI (November 20, 2026), SFCompute had a median time to interactive of 621ms versus 2.28s for Tensorlake, 3.66x faster. In DAX (November 20, 2026), SFCompute had a median total workload time of 39.91s versus 219.25s for Tensorlake, 5.49x faster.

Both providers have results in 2 sandbox benchmarks: Burst TTI and DAX. All of them have a speed verdict.

Success rates in the latest runs: Burst TTI: SFCompute 100%, Tensorlake 100%; DAX: SFCompute 100%, Tensorlake 100%.

Burst TTI and DAX results are from November 20, 2026.