llamaperf
Submit

StepFun 3.7

StepFun · 2 reports

StepFun 3.7 198B · 11B active at 4-bit generates a median of 33.9 tokens per second on the M5 Max 128GB, the hardware it is most run on, from one plain run (community reports, as of 26 Oct 2026).

As of 26 Oct 2026, StepFun 3.7 198B · 11B active at 4-bit on the hardware it is most run on, with the median of plain runs (one device, one request, no speculative decoding, the whole model in its memory):

StepFun 3.7 VRAM requirements by size and quant →

How does StepFun 3.7 run on your hardware?

Pick your GPU or Mac, then read the speeds people reported for StepFun 3.7 on it, or estimate memory fit and speed.

Buying hardware? Compare GPUs →

Free to use. No account needed. Memory estimates and community measurements are labelled separately.

Run StepFun 3.7 yourself? Add your numbers.

Your coding agent can run a speed test on your machine and send it here with a free account. Or paste a result you've already posted.

Benchmark with your agentPaste a result

Reported performance for StepFun 3.7

Filter this model’s reports by setup →
Thin page (2 of 3 reports needed for indexing). Add yours.

StepFun 3.7 Step 3.7 Flash / IQ4_XS / ROCm b9843 native PP2080 d512

AMD Radeon 8060S integrated GPU (Strix Halo; system RAM unreported) · llama.cpp b9843 (86b94708f) / stock ROCm

prompt processing (prefill):
288.8 tokens/s
quant:
IQ4_XS (GGUF)
kv:
F16
flash attention:
on

Reported by the source; GPU count, offloading and concurrent requests can change these figures. Prompt-processing speed is input throughput, not time to first token. Check the full setup before comparing.

User reports StepFun 3.7 at 288.8 t/s prompt processing on an AMD Ryzen AI MAX+ 395 with Radeon 8060S graphics. Setup is llama.cpp b9843 with IQ4_XS, f16 K/V cache and flash attention, one participating Radeon via ROCm with ngl=999. The 288.8 t/s figure is the stock batch2048/ubatch512 arm at depth 512, measured over three repeats; the same arm falls to 146.025623 t/s at depth 100000.

Oct 13, 2026
Apple hardware
generation:
33.9 tokens/s
quant:
Q4_K_S (GGUF)

Reported by the source; GPU count, offloading and concurrent requests can change these figures. Prompt-processing speed is shown only when the source reports it; generation speed cannot tell us time to first token. Check the full setup before comparing.

User benchmarks prompt processing at 65,536 context, with throughput falling from 62.80 t/s at 128 context to 33.92 t/s at 65,536. Memory peaked at ~120+ GB, and the system stayed sluggish but usable.

Aug 28, 2026

Get a weekly email of new StepFun 3.7 reports on any GPU.

Email me new reports

Reported configurations

Individual observations from this page, not expected speeds or a ranking. Select a model to open its report, with the source and offloading conditions. Reported context may be a limit rather than actual input length.

Model / hardwareQuant / engineReported contextGeneration
StepFun 3.7 198B (11B active) Flash
M5 Max 128GB
Q4_K_S
llama.cpp
65,53633.9 tokens/s

Frequently asked

What hardware do people run StepFun 3.7 on?

The hardware StepFun 3.7 198B · 11B active at 4-bit is most run on: M5 Max 128GB (33.9 t/s median, one run). Medians of plain runs, as of 26 Oct 2026.

Which inference engines do people use for StepFun 3.7?

Reports of StepFun 3.7 name llama.cpp (1 report), as of 26 Oct 2026.