AMD · consumer

Radeon RX 6500 XT

Radeon RX 6500 XT has 4 GB of VRAM at 144 GB/s — about 3.72 GiB usable after driver and compositor overhead. 774 of 2118 indexed models fit at 4K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
4 GB
GDDR6
Bandwidth
144 GB/s
64-bit bus
Tensor FP16
—
dense
TDP
107 W
$199 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 637vision language 54embedding 25audio asr 37audio tts 19video 2

What fits at 4K context

largest quantization that fits, per model · 774 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
granite-3.3-2b-instructQ8_02.5B2.51 GiB0.31 GiB3.72 GiB0.00 GiB32±26.5%
granite-3.2-2b-instructQ8_02.5B2.51 GiB0.31 GiB3.72 GiB0.00 GiB32±26.5%
granite-vision-3.2-2bQ8_03.0B2.51 GiB0.31 GiB3.72 GiB0.00 GiB32±26.5%
Ministral-3-8B-Instruct-2512UD-IQ1_M8.9B2.25 GiB0.53 GiB3.72 GiB0.00 GiB32±26.5%
CycleGRPO-4BI1-Q3_K_L4.8B2.24 GiB0.56 GiB3.72 GiB0.00 GiB32±26.5%
gemma-4-E2B-it-ultra-uncensored-hereticIQ4_XS5.1B2.77 GiB0.05 GiB3.72 GiB0.00 GiB32±26.5%
Jan-v3-4B-base-instructQ3_K_L4.4B2.24 GiB0.56 GiB3.72 GiB0.00 GiB32±26.5%
Jan-code-4bQ3_K_L4.4B2.24 GiB0.56 GiB3.72 GiB0.00 GiB32±26.5%
Qwen3-4B-BaseQ3_K_L4.0B2.24 GiB0.56 GiB3.71 GiB0.01 GiB32±26.5%
granite-3.3-8b-instructUD-IQ2_XXS8.2B2.16 GiB0.63 GiB3.71 GiB0.01 GiB32±26.5%
Ministral-3-3B-Instruct-2512-BF16Q5_K_L4.3B2.39 GiB0.41 GiB3.71 GiB0.01 GiB32±26.5%
Llama-3.2-3B-Instruct-abliteratedI1-Q5_K_S3.6B2.37 GiB0.44 GiB3.71 GiB0.01 GiB32±26.5%
Llama-3.2-3B-Instruct-uncensoredQ5_K_S3.6B2.37 GiB0.44 GiB3.71 GiB0.01 GiB32±26.5%
NVIDIA-Nemotron-3-Nano-4B-BF16IQ2_M4.0B2.13 GiB0.66 GiB3.71 GiB0.01 GiB32±26.5%
Ministral-3-8B-Reasoning-2512UD-IQ1_M8.9B2.24 GiB0.53 GiB3.71 GiB0.01 GiB32±26.5%
Nemotron-Mini-4B-InstructIQ4_XS4.2B2.29 GiB0.50 GiB3.71 GiB0.01 GiB32±26.5%
gemma-3n-E2B-itIQ4_XS5.4B2.71 GiB0.09 GiB3.71 GiB0.01 GiB32±26.5%
Hubble-4B-v1Q3_K_L4.5B2.30 GiB0.50 GiB3.71 GiB0.01 GiB32±26.5%
Aura-4BI1-Q3_K_L4.5B2.30 GiB0.50 GiB3.71 GiB0.01 GiB32±26.5%
magnum-v2-4bI1-Q3_K_L4.5B2.30 GiB0.50 GiB3.71 GiB0.01 GiB32±26.5%
Impish_LLAMA_4BQ3_K_L4.5B2.30 GiB0.50 GiB3.71 GiB0.01 GiB32±26.5%
Llama-3.1-Minitron-4B-Width-BaseQ3_K_L4.5B2.30 GiB0.50 GiB3.71 GiB0.01 GiB32±26.5%
Holo-3.1-4BIQ4_XS5.2B2.67 GiB0.13 GiB3.71 GiB0.01 GiB32±26.5%
ToriiGate-0.5IQ4_XS5.2B2.67 GiB0.13 GiB3.71 GiB0.01 GiB32±26.5%
Yi-6B-ChatI1-IQ3_S6.1B2.53 GiB0.25 GiB3.71 GiB0.01 GiB32±26.5%
Hunyuan-7B-InstructIQ2_XS7.5B2.26 GiB0.50 GiB3.70 GiB0.02 GiB32±26.5%
dolphin-2_6-phi-2Q8_02.8B2.75 GiB0.00 GiB3.70 GiB0.02 GiB32±26.5%
Yi-1.5-6B-ChatQ3_K_S6.1B2.52 GiB0.25 GiB3.70 GiB0.02 GiB32±26.5%
LFM2.5-Queen-Opus-4.7-8B-A1BMoEI1-Q2_K_S8.5B2.75 GiB0.05 GiB3.70 GiB0.02 GiB87±37%
LFM2.5-8B-A1B-KO-SFTMoEI1-Q2_K_S8.5B2.75 GiB0.05 GiB3.70 GiB0.02 GiB87±37%
LFM2.5-8B-A1B-hereticMoEI1-Q2_K_S8.5B2.75 GiB0.05 GiB3.70 GiB0.02 GiB87±37%
LFM2.5-8B-A1B-SOMPOA-heresyMoEI1-Q2_K_S8.5B2.75 GiB0.05 GiB3.70 GiB0.02 GiB87±37%
Huihui-LFM2.5-8B-A1B-abliteratedMoEI1-Q2_K_S8.5B2.75 GiB0.05 GiB3.70 GiB0.02 GiB87±37%
Supertron2.1-8B-A1BMoEI1-Q2_K_S8.5B2.75 GiB0.05 GiB3.70 GiB0.02 GiB87±37%
Luna-7B-A4BMoEI1-IQ2_M6.7B2.22 GiB0.56 GiB3.70 GiB0.02 GiB29±37%
AfriqueQwen3.5-4BI1-IQ4_XS5.2B2.66 GiB0.13 GiB3.69 GiB0.03 GiB32±26.5%
TimeOmni-1-4BI1-IQ4_XS5.2B2.66 GiB0.13 GiB3.69 GiB0.03 GiB32±26.5%
Qwen3-VL-4B-Instruct-Unredacted-MAXI1-Q4_K_S4.4B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Qwen3-VL-4B-ThinkingQ4_K_S4.4B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Qwen3-VL-4B-Thinking-Unredacted-MAXI1-Q4_K_S4.4B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Zubr1.0-VL-4BI1-Q4_K_S4.4B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Qwen3-VL-4B-Instruct-Uncensored-abliteratedQ4_K_S4.4B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
PopiT-Qwen3-4B-Medical-SFT-1128Q4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Huihui-Qwen3-VL-4B-Instruct-abliteratedI1-Q4_K_S4.4B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Qwen3-VL-4B-Instruct-UncensoredI1-Q4_K_S4.4B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Qwen3-VL-4B-InstructQ4_K_S4.4B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
OpenCaption-4B-VL-SFT-v1.0I1-Q4_K_S4.4B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Parable-Qwen3-4B-Claude-Fable-5I1-Q4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Logics-Parsing-v2Q4_K_S4.4B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Jan-v1-4BQ4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Qwen3-4b-Z-Image-Turbo-AbliteratedV1I1-Q4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Z-Image-Engineer-V6Q4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Darwin-4B-ChimeraI1-Q4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Qwen3-4BQ4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Huihui-Qwen3-4B-Instruct-2507-abliteratedQ4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Josiefied-Qwen3-4B-abliterated-v2Q4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Qwen3-4B-Thinking-2507Q4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Jan-nano-128kQ4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Qwen3-4B-Instruct-2507Q4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
Jan-nanoQ4_K_S4.0B2.22 GiB0.56 GiB3.69 GiB0.03 GiB32±26.5%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Questions people ask

What AI models can a Radeon RX 6500 XT run?
774 of 2118 indexed open-weight models fit a Radeon RX 6500 XT at 4,096 context with f16 KV cache, the largest being granite-3.3-2b-instruct at Q8_0. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon RX 6500 XT actually have?
Its nameplate is 4 GB, but about 3.72 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Radeon RX 6500 XT fast for local AI?
Its memory bandwidth is 144 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.