Nanbeige4.1-3B

Nanbeige · nanbeige/nanbeige4-1-3b

ProviderNanbeige
Familynanbeige
Typellm-chat
Statusactive
Released2026-02-10
Updated2026-03-25
Parameters3.9B
Open weightsyes

Lineage

RelationshipModel
Is a finetune ofNanbeige/Nanbeige4-3B-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/sbf167.87 GB~1868499.8~7473999.0 q4
NVIDIA GB200 Grace Blackwell Superchip372.0 GB16000.0 GB/sbf167.87 GB~1423.6~5694.5 q4
AMD Instinct MI355X288.0 GB8000.0 GB/sbf167.87 GB~711.8~2847.2 q4
NVIDIA B200180.0 GB7700.0 GB/sbf167.87 GB~685.1~2740.5 q4
Google TPU7x (Ironwood)192.0 GB7380.0 GB/sbf167.87 GB~656.6~2626.6 q4
AMD Instinct MI325X256.0 GB6000.0 GB/sbf167.87 GB~533.9~2135.4 q4
AMD Instinct MI300X192.0 GB5300.0 GB/sbf167.87 GB~471.6~1886.3 q4
NVIDIA H200 SXM141.0 GB4800.0 GB/sbf167.87 GB~427.1~1708.3 q4
NVIDIA H100 NVL94.0 GB3900.0 GB/sbf167.87 GB~347.0~1388.0 q4
NVIDIA H100 SXM80.0 GB3350.0 GB/sbf167.87 GB~298.1~1192.3 q4
AMD Instinct MI250X128.0 GB3200.0 GB/sbf167.87 GB~284.7~1138.9 q4
Google TPU v5p95.0 GB2765.0 GB/sbf167.87 GB~246.0~984.1 q4
NVIDIA A100 80GB SXM80.0 GB2039.0 GB/sbf167.87 GB~181.4~725.7 q4
NVIDIA H100 PCIe80.0 GB2000.0 GB/sbf167.87 GB~178.0~711.8 q4
NVIDIA GeForce RTX 509032.0 GB1792.0 GB/sbf167.87 GB~159.4~637.8 q4
Google TPU v6e (Trillium)32.0 GB1638.0 GB/sbf167.87 GB~145.7~583.0 q4
AMD Instinct MI21064.0 GB1600.0 GB/sbf167.87 GB~142.4~569.4 q4
NVIDIA A100 40GB SXM40.0 GB1555.0 GB/sbf167.87 GB~138.4~553.4 q4
Google TPU v432.0 GB1200.0 GB/sbf167.87 GB~106.8~427.1 q4
NVIDIA GeForce RTX 409024.0 GB1008.0 GB/sbf167.87 GB~89.7~358.8 q4
AMD Radeon RX 7900 XTX24.0 GB960.0 GB/sbf167.87 GB~85.4~341.7 q4
NVIDIA GeForce RTX 508016.0 GB960.0 GB/sbf167.87 GB~85.4~341.7 q4
NVIDIA GeForce RTX 309024.0 GB936.0 GB/sbf167.87 GB~83.3~333.1 q4
NVIDIA L40S48.0 GB864.0 GB/sbf167.87 GB~76.9~307.5 q4
Apple M3 Ultra512.0 GB819.0 GB/sbf167.87 GB~72.9~291.5 q4
AMD Radeon RX 7900 XT20.0 GB800.0 GB/sbf167.87 GB~71.2~284.7 q4
Apple M2 Ultra192.0 GB800.0 GB/sbf167.87 GB~71.2~284.7 q4
Google TPU v5e16.0 GB800.0 GB/sbf167.87 GB~71.2~284.7 q4
NVIDIA GeForce RTX 4080 SUPER16.0 GB736.0 GB/sbf167.87 GB~65.5~261.9 q4
NVIDIA GeForce RTX 4070 Ti SUPER16.0 GB672.0 GB/sbf167.87 GB~59.8~239.2 q4
AMD Radeon RX 9070 XT16.0 GB640.0 GB/sbf167.87 GB~56.9~227.8 q4
Apple M4 Max128.0 GB546.0 GB/sbf167.87 GB~48.6~194.3 q4
Apple M1 Max64.0 GB400.0 GB/sbf167.87 GB~35.6~142.4 q4
Apple M2 Max96.0 GB400.0 GB/sbf167.87 GB~35.6~142.4 q4
Apple M3 Max128.0 GB400.0 GB/sbf167.87 GB~35.6~142.4 q4
NVIDIA GeForce RTX 3060 12GB12.0 GB360.0 GB/sbf167.87 GB~32.0~128.1 q4
NVIDIA L424.0 GB300.0 GB/sbf167.87 GB~26.7~106.8 q4
NVIDIA GeForce RTX 4060 Ti 16GB16.0 GB288.0 GB/sbf167.87 GB~25.6~102.5 q4
Apple M4 Pro64.0 GB273.0 GB/sbf167.87 GB~24.3~97.2 q4
Apple M432.0 GB120.0 GB/sbf167.87 GB~10.7~42.7 q4

Where it runs

Taken from the card's availability section.

PlatformId on that platform
Hugging FaceNanbeige/Nanbeige4.1-3B

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_lcrNanbeige4.1-3B0.0%2026-09-10 evaluatedindependent evaluatorsource
critptNanbeige4.1-3B0.0%2026-09-10 evaluatedindependent evaluatorsource
gdpval_aaNanbeige4.1-3B0.0%2026-09-10 evaluatedindependent evaluatorsource
gpqa_diamondNanbeige4.1-3B84.95%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