MiniCPM5-2B

OpenBMB · openbmb/minicpm5-2b

ProviderOpenBMB
Familyminicpm
Typellm-chat
Statusactive
Released2026-09-06
Updated2026-09-10
Parameters2.5B
Open weightsyes

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/sbf165.03 GB~2920425.6~11681702.3 q4
NVIDIA GB200 Grace Blackwell Superchip372.0 GB16000.0 GB/sbf165.03 GB~2225.1~8900.3 q4
AMD Instinct MI355X288.0 GB8000.0 GB/sbf165.03 GB~1112.5~4450.2 q4
NVIDIA B200180.0 GB7700.0 GB/sbf165.03 GB~1070.8~4283.3 q4
Google TPU7x (Ironwood)192.0 GB7380.0 GB/sbf165.03 GB~1026.3~4105.3 q4
AMD Instinct MI325X256.0 GB6000.0 GB/sbf165.03 GB~834.4~3337.6 q4
AMD Instinct MI300X192.0 GB5300.0 GB/sbf165.03 GB~737.1~2948.2 q4
NVIDIA H200 SXM141.0 GB4800.0 GB/sbf165.03 GB~667.5~2670.1 q4
NVIDIA H100 NVL94.0 GB3900.0 GB/sbf165.03 GB~542.4~2169.5 q4
NVIDIA H100 SXM80.0 GB3350.0 GB/sbf165.03 GB~465.9~1863.5 q4
AMD Instinct MI250X128.0 GB3200.0 GB/sbf165.03 GB~445.0~1780.1 q4
Google TPU v5p95.0 GB2765.0 GB/sbf165.03 GB~384.5~1538.1 q4
NVIDIA A100 80GB SXM80.0 GB2039.0 GB/sbf165.03 GB~283.6~1134.2 q4
NVIDIA H100 PCIe80.0 GB2000.0 GB/sbf165.03 GB~278.1~1112.5 q4
NVIDIA GeForce RTX 509032.0 GB1792.0 GB/sbf165.03 GB~249.2~996.8 q4
Google TPU v6e (Trillium)32.0 GB1638.0 GB/sbf165.03 GB~227.8~911.2 q4
AMD Instinct MI21064.0 GB1600.0 GB/sbf165.03 GB~222.5~890.0 q4
NVIDIA A100 40GB SXM40.0 GB1555.0 GB/sbf165.03 GB~216.3~865.0 q4
Google TPU v432.0 GB1200.0 GB/sbf165.03 GB~166.9~667.5 q4
NVIDIA GeForce RTX 409024.0 GB1008.0 GB/sbf165.03 GB~140.2~560.7 q4
AMD Radeon RX 7900 XTX24.0 GB960.0 GB/sbf165.03 GB~133.5~534.0 q4
NVIDIA GeForce RTX 508016.0 GB960.0 GB/sbf165.03 GB~133.5~534.0 q4
NVIDIA GeForce RTX 309024.0 GB936.0 GB/sbf165.03 GB~130.2~520.7 q4
NVIDIA L40S48.0 GB864.0 GB/sbf165.03 GB~120.2~480.6 q4
Apple M3 Ultra512.0 GB819.0 GB/sbf165.03 GB~113.9~455.6 q4
AMD Radeon RX 7900 XT20.0 GB800.0 GB/sbf165.03 GB~111.3~445.0 q4
Apple M2 Ultra192.0 GB800.0 GB/sbf165.03 GB~111.3~445.0 q4
Google TPU v5e16.0 GB800.0 GB/sbf165.03 GB~111.3~445.0 q4
NVIDIA GeForce RTX 4080 SUPER16.0 GB736.0 GB/sbf165.03 GB~102.4~409.4 q4
NVIDIA GeForce RTX 4070 Ti SUPER16.0 GB672.0 GB/sbf165.03 GB~93.5~373.8 q4
AMD Radeon RX 9070 XT16.0 GB640.0 GB/sbf165.03 GB~89.0~356.0 q4
Apple M4 Max128.0 GB546.0 GB/sbf165.03 GB~75.9~303.7 q4
Apple M1 Max64.0 GB400.0 GB/sbf165.03 GB~55.6~222.5 q4
Apple M2 Max96.0 GB400.0 GB/sbf165.03 GB~55.6~222.5 q4
Apple M3 Max128.0 GB400.0 GB/sbf165.03 GB~55.6~222.5 q4
NVIDIA GeForce RTX 3060 12GB12.0 GB360.0 GB/sbf165.03 GB~50.1~200.3 q4
NVIDIA L424.0 GB300.0 GB/sbf165.03 GB~41.7~166.9 q4
NVIDIA GeForce RTX 4060 Ti 16GB16.0 GB288.0 GB/sbf165.03 GB~40.1~160.2 q4
Apple M4 Pro64.0 GB273.0 GB/sbf165.03 GB~38.0~151.9 q4
Apple M432.0 GB120.0 GB/sbf165.03 GB~16.7~66.8 q4

Where it runs

Taken from the card's availability section.

PlatformId on that platform
Hugging Faceopenbmb/MiniCPM5-2B

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_lcrMiniCPM5-2B59.33%2026-09-10 evaluatedindependent evaluatorsource
critptMiniCPM5-2B0.29%2026-09-10 evaluatedindependent evaluatorsource
gdpval_aaMiniCPM5-2B16.38%2026-09-10 evaluatedindependent evaluatorsource
gpqa_diamondMiniCPM5-2B70.2%2026-09-10 evaluatedindependent evaluatorsource
scicodeMiniCPM5-2B26.27%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