Granite 4.1 3B

IBM · ibm/granite-4-1-3b

ProviderIBM
Familygranite
Typellm-chat
Statusactive
Released2026-04-06
Updated2026-05-04
Parameters3.4B
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/sbf166.81 GB~2159962.7~8639851.0 q4
NVIDIA GB200 Grace Blackwell Superchip372.0 GB16000.0 GB/sbf166.81 GB~1645.7~6582.7 q4
AMD Instinct MI355X288.0 GB8000.0 GB/sbf166.81 GB~822.8~3291.4 q4
NVIDIA B200180.0 GB7700.0 GB/sbf166.81 GB~792.0~3167.9 q4
Google TPU7x (Ironwood)192.0 GB7380.0 GB/sbf166.81 GB~759.1~3036.3 q4
AMD Instinct MI325X256.0 GB6000.0 GB/sbf166.81 GB~617.1~2468.5 q4
AMD Instinct MI300X192.0 GB5300.0 GB/sbf166.81 GB~545.1~2180.5 q4
NVIDIA H200 SXM141.0 GB4800.0 GB/sbf166.81 GB~493.7~1974.8 q4
NVIDIA H100 NVL94.0 GB3900.0 GB/sbf166.81 GB~401.1~1604.5 q4
NVIDIA H100 SXM80.0 GB3350.0 GB/sbf166.81 GB~344.6~1378.3 q4
AMD Instinct MI250X128.0 GB3200.0 GB/sbf166.81 GB~329.1~1316.5 q4
Google TPU v5p95.0 GB2765.0 GB/sbf166.81 GB~284.4~1137.6 q4
NVIDIA A100 80GB SXM80.0 GB2039.0 GB/sbf166.81 GB~209.7~838.9 q4
NVIDIA H100 PCIe80.0 GB2000.0 GB/sbf166.81 GB~205.7~822.8 q4
NVIDIA GeForce RTX 509032.0 GB1792.0 GB/sbf166.81 GB~184.3~737.3 q4
Google TPU v6e (Trillium)32.0 GB1638.0 GB/sbf166.81 GB~168.5~673.9 q4
AMD Instinct MI21064.0 GB1600.0 GB/sbf166.81 GB~164.6~658.3 q4
NVIDIA A100 40GB SXM40.0 GB1555.0 GB/sbf166.81 GB~159.9~639.8 q4
Google TPU v432.0 GB1200.0 GB/sbf166.81 GB~123.4~493.7 q4
NVIDIA GeForce RTX 409024.0 GB1008.0 GB/sbf166.81 GB~103.7~414.7 q4
AMD Radeon RX 7900 XTX24.0 GB960.0 GB/sbf166.81 GB~98.7~395.0 q4
NVIDIA GeForce RTX 508016.0 GB960.0 GB/sbf166.81 GB~98.7~395.0 q4
NVIDIA GeForce RTX 309024.0 GB936.0 GB/sbf166.81 GB~96.3~385.1 q4
NVIDIA L40S48.0 GB864.0 GB/sbf166.81 GB~88.9~355.5 q4
Apple M3 Ultra512.0 GB819.0 GB/sbf166.81 GB~84.2~337.0 q4
AMD Radeon RX 7900 XT20.0 GB800.0 GB/sbf166.81 GB~82.3~329.1 q4
Apple M2 Ultra192.0 GB800.0 GB/sbf166.81 GB~82.3~329.1 q4
Google TPU v5e16.0 GB800.0 GB/sbf166.81 GB~82.3~329.1 q4
NVIDIA GeForce RTX 4080 SUPER16.0 GB736.0 GB/sbf166.81 GB~75.7~302.8 q4
NVIDIA GeForce RTX 4070 Ti SUPER16.0 GB672.0 GB/sbf166.81 GB~69.1~276.5 q4
AMD Radeon RX 9070 XT16.0 GB640.0 GB/sbf166.81 GB~65.8~263.3 q4
Apple M4 Max128.0 GB546.0 GB/sbf166.81 GB~56.2~224.6 q4
Apple M1 Max64.0 GB400.0 GB/sbf166.81 GB~41.1~164.6 q4
Apple M2 Max96.0 GB400.0 GB/sbf166.81 GB~41.1~164.6 q4
Apple M3 Max128.0 GB400.0 GB/sbf166.81 GB~41.1~164.6 q4
NVIDIA GeForce RTX 3060 12GB12.0 GB360.0 GB/sbf166.81 GB~37.0~148.1 q4
NVIDIA L424.0 GB300.0 GB/sbf166.81 GB~30.9~123.4 q4
NVIDIA GeForce RTX 4060 Ti 16GB16.0 GB288.0 GB/sbf166.81 GB~29.6~118.5 q4
Apple M4 Pro64.0 GB273.0 GB/sbf166.81 GB~28.1~112.3 q4
Apple M432.0 GB120.0 GB/sbf166.81 GB~12.3~49.4 q4

Where it runs

Taken from the card's availability section.

PlatformId on that platform
Hugging Faceibm-granite/granite-4.1-3b

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_lcrGranite 4.1 3B6.0%2026-09-10 evaluatedindependent evaluatorsource
critptGranite 4.1 3B0.0%2026-09-10 evaluatedindependent evaluatorsource
gdpval_aaGranite 4.1 3B0.0%2026-09-10 evaluatedindependent evaluatorsource
gpqa_diamondGranite 4.1 3B31.41%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