G9v3-39A5B

AI9Stars · ai9stars/g9v3-39a5b

ProviderAI9Stars
Familyg9
Typellm-chat
Statusactive
Released2026-07-21
Updated2026-08-19
Parameters39B
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/sq629.23 GB~3822400.6~5733600.9 q4
NVIDIA GB200 Grace Blackwell Superchip372.0 GB16000.0 GB/sbf1677.93 GB~1092.1~4368.5 q4
AMD Instinct MI355X288.0 GB8000.0 GB/sbf1677.93 GB~546.1~2184.2 q4
NVIDIA B200180.0 GB7700.0 GB/sbf1677.93 GB~525.6~2102.3 q4
Google TPU7x (Ironwood)192.0 GB7380.0 GB/sbf1677.93 GB~503.7~2015.0 q4
AMD Instinct MI325X256.0 GB6000.0 GB/sbf1677.93 GB~409.5~1638.2 q4
AMD Instinct MI300X192.0 GB5300.0 GB/sbf1677.93 GB~361.8~1447.1 q4
NVIDIA H200 SXM141.0 GB4800.0 GB/sbf1677.93 GB~327.6~1310.5 q4
NVIDIA H100 NVL94.0 GB3900.0 GB/sfp838.97 GB~532.4~1064.8 q4
NVIDIA H100 SXM80.0 GB3350.0 GB/sfp838.97 GB~457.3~914.6 q4
AMD Instinct MI250X128.0 GB3200.0 GB/sbf1677.93 GB~218.4~873.7 q4
Google TPU v5p95.0 GB2765.0 GB/sfp838.97 GB~377.5~754.9 q4
NVIDIA A100 80GB SXM80.0 GB2039.0 GB/sfp838.97 GB~278.4~556.7 q4
NVIDIA H100 PCIe80.0 GB2000.0 GB/sfp838.97 GB~273.0~546.1 q4
NVIDIA GeForce RTX 509032.0 GB1792.0 GB/sq419.48 GB~489.3~489.3 q4
Google TPU v6e (Trillium)32.0 GB1638.0 GB/sq419.48 GB~447.2~447.2 q4
AMD Instinct MI21064.0 GB1600.0 GB/sfp838.97 GB~218.4~436.8 q4
NVIDIA A100 40GB SXM40.0 GB1555.0 GB/sq629.23 GB~283.0~424.6 q4
Google TPU v432.0 GB1200.0 GB/sq419.48 GB~327.6~327.6 q4
NVIDIA L40S48.0 GB864.0 GB/sq629.23 GB~157.3~235.9 q4
Apple M3 Ultra512.0 GB819.0 GB/sbf1677.93 GB~55.9~223.6 q4
Apple M2 Ultra192.0 GB800.0 GB/sbf1677.93 GB~54.6~218.4 q4
Apple M4 Max128.0 GB546.0 GB/sbf1677.93 GB~37.3~149.1 q4
Apple M1 Max64.0 GB400.0 GB/sfp838.97 GB~54.6~109.2 q4
Apple M2 Max96.0 GB400.0 GB/sfp838.97 GB~54.6~109.2 q4
Apple M3 Max128.0 GB400.0 GB/sbf1677.93 GB~27.3~109.2 q4
Apple M4 Pro64.0 GB273.0 GB/sfp838.97 GB~37.3~74.5 q4
Apple M432.0 GB120.0 GB/sq419.48 GB~32.8~32.8 q4

Where it runs

Taken from the card's availability section.

PlatformId on that platform
Hugging Faceai9stars/G9v3-39A5B

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_lcrG9v3-39A5B65.33%2026-09-10 evaluatedindependent evaluatorsource
critptG9v3-39A5B0.29%2026-09-10 evaluatedindependent evaluatorsource
gdpval_aaG9v3-39A5B34.44%2026-09-10 evaluatedindependent evaluatorsource
gpqa_diamondG9v3-39A5B80.51%2026-09-10 evaluatedindependent evaluatorsource
scicodeG9v3-39A5B36.81%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