Step3 VL 10B

StepFun · stepfun/step3-vl-10b

ProviderStepFun
Familystep
Typevlm
Statusactive
Released2026-01-13
Updated2026-02-04
Parameters10.2B
Open weightsyes

Lineage

RelationshipModel
Is a finetune ofstepfun-ai/Step3-VL-10B-Base

What it fits on

Ordered by predicted speed. Bandwidth sets decode rate; memory decides whether it runs at all.

Every figure here is computed, not measured. Fit is weights at each quantisation against device memory, with a 25% allowance for the KV cache, activations and the OS. Decode rate is the memory-bandwidth roofline at 70% efficiency. Nobody has run this model on these devices.
DeviceMemoryBandwidthBest qualityWeightstok/sFastest
Cerebras WSE-344.0 GB21000000.0 GB/sbf1620.34 GB~722589.5~2890358.1 q4
NVIDIA GB200 Grace Blackwell Superchip372.0 GB16000.0 GB/sbf1620.34 GB~550.5~2202.2 q4
AMD Instinct MI355X288.0 GB8000.0 GB/sbf1620.34 GB~275.3~1101.1 q4
NVIDIA B200180.0 GB7700.0 GB/sbf1620.34 GB~264.9~1059.8 q4
Google TPU7x (Ironwood)192.0 GB7380.0 GB/sbf1620.34 GB~253.9~1015.8 q4
AMD Instinct MI325X256.0 GB6000.0 GB/sbf1620.34 GB~206.5~825.8 q4
AMD Instinct MI300X192.0 GB5300.0 GB/sbf1620.34 GB~182.4~729.5 q4
NVIDIA H200 SXM141.0 GB4800.0 GB/sbf1620.34 GB~165.2~660.7 q4
NVIDIA H100 NVL94.0 GB3900.0 GB/sbf1620.34 GB~134.2~536.8 q4
NVIDIA H100 SXM80.0 GB3350.0 GB/sbf1620.34 GB~115.3~461.1 q4
AMD Instinct MI250X128.0 GB3200.0 GB/sbf1620.34 GB~110.1~440.4 q4
Google TPU v5p95.0 GB2765.0 GB/sbf1620.34 GB~95.1~380.6 q4
NVIDIA A100 80GB SXM80.0 GB2039.0 GB/sbf1620.34 GB~70.2~280.6 q4
NVIDIA H100 PCIe80.0 GB2000.0 GB/sbf1620.34 GB~68.8~275.3 q4
NVIDIA GeForce RTX 509032.0 GB1792.0 GB/sbf1620.34 GB~61.7~246.6 q4
Google TPU v6e (Trillium)32.0 GB1638.0 GB/sbf1620.34 GB~56.4~225.4 q4
AMD Instinct MI21064.0 GB1600.0 GB/sbf1620.34 GB~55.1~220.2 q4
NVIDIA A100 40GB SXM40.0 GB1555.0 GB/sbf1620.34 GB~53.5~214.0 q4
Google TPU v432.0 GB1200.0 GB/sbf1620.34 GB~41.3~165.2 q4
NVIDIA GeForce RTX 409024.0 GB1008.0 GB/sfp810.17 GB~69.4~138.7 q4
AMD Radeon RX 7900 XTX24.0 GB960.0 GB/sfp810.17 GB~66.1~132.1 q4
NVIDIA GeForce RTX 508016.0 GB960.0 GB/sfp810.17 GB~66.1~132.1 q4
NVIDIA GeForce RTX 309024.0 GB936.0 GB/sfp810.17 GB~64.4~128.8 q4
NVIDIA L40S48.0 GB864.0 GB/sbf1620.34 GB~29.7~118.9 q4
Apple M3 Ultra512.0 GB819.0 GB/sbf1620.34 GB~28.2~112.7 q4
AMD Radeon RX 7900 XT20.0 GB800.0 GB/sfp810.17 GB~55.1~110.1 q4
Apple M2 Ultra192.0 GB800.0 GB/sbf1620.34 GB~27.5~110.1 q4
Google TPU v5e16.0 GB800.0 GB/sfp810.17 GB~55.1~110.1 q4
NVIDIA GeForce RTX 4080 SUPER16.0 GB736.0 GB/sfp810.17 GB~50.7~101.3 q4
NVIDIA GeForce RTX 4070 Ti SUPER16.0 GB672.0 GB/sfp810.17 GB~46.2~92.5 q4
AMD Radeon RX 9070 XT16.0 GB640.0 GB/sfp810.17 GB~44.0~88.1 q4
Apple M4 Max128.0 GB546.0 GB/sbf1620.34 GB~18.8~75.1 q4
Apple M1 Max64.0 GB400.0 GB/sbf1620.34 GB~13.8~55.1 q4
Apple M2 Max96.0 GB400.0 GB/sbf1620.34 GB~13.8~55.1 q4
Apple M3 Max128.0 GB400.0 GB/sbf1620.34 GB~13.8~55.1 q4
NVIDIA GeForce RTX 3060 12GB12.0 GB360.0 GB/sq67.63 GB~33.0~49.5 q4
NVIDIA L424.0 GB300.0 GB/sfp810.17 GB~20.6~41.3 q4
NVIDIA GeForce RTX 4060 Ti 16GB16.0 GB288.0 GB/sfp810.17 GB~19.8~39.6 q4
Apple M4 Pro64.0 GB273.0 GB/sbf1620.34 GB~9.4~37.6 q4
Apple M432.0 GB120.0 GB/sbf1620.34 GB~4.1~16.5 q4

Where it runs

Taken from the card's availability section.

PlatformId on that platform
Hugging Facestepfun-ai/Step3-VL-10B

Verified benchmark evidence

Each row was checked against its source by a reviewer, and carries the model identifier as actually evaluated — which is not always the same as this card's.
BenchmarkEvaluated asScoreEvidence dateSource kind
aa_lcrStep3 VL 10B0.0%2026-09-10 evaluatedindependent evaluatorsource
critptStep3 VL 10B0.0%2026-09-10 evaluatedindependent evaluatorsource
gpqa_diamondStep3 VL 10B68.99%2026-09-10 evaluatedindependent evaluatorsource

Reported benchmark scores

This card reports no benchmark scores yet.

BenchmarkCatalogue standingScore

Data

This card as JSON · See it in the graph · Edit on GitHub