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. 342 of 2118 indexed models fit at 128K context with q4_0 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 252vision language 29audio tts 17video 2embedding 13audio asr 29

What fits at 128K context

largest quantization that fits, per model · 342 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Trinity-Nano-PreviewMoEQ2_K_L6.1B2.32 GiB0.52 GiB3.72 GiB0.00 GiB61±37%
GLM-OCRI1-Q5_K_S1.3B0.59 GiB2.25 GiB3.72 GiB0.00 GiB32±26.5%
LFM2.5-Queen-Opus-4.7-8B-A1BMoEI1-IQ2_XS8.5B2.40 GiB0.42 GiB3.72 GiB0.00 GiB58±37%
LFM2.5-8B-A1B-KO-SFTMoEI1-IQ2_XS8.5B2.40 GiB0.42 GiB3.72 GiB0.00 GiB58±37%
LFM2.5-8B-A1B-hereticMoEI1-IQ2_XS8.5B2.40 GiB0.42 GiB3.72 GiB0.00 GiB58±37%
LFM2.5-8B-A1B-SOMPOA-heresyMoEI1-IQ2_XS8.5B2.40 GiB0.42 GiB3.72 GiB0.00 GiB58±37%
Huihui-LFM2.5-8B-A1B-abliteratedMoEI1-IQ2_XS8.5B2.40 GiB0.42 GiB3.72 GiB0.00 GiB58±37%
Supertron2.1-8B-A1BMoEI1-IQ2_XS8.5B2.40 GiB0.42 GiB3.72 GiB0.00 GiB58±37%
LFM2.5-8B-A1BMoEUD-IQ1_M8.5B2.40 GiB0.42 GiB3.72 GiB0.00 GiB58±37%
TinyLlama-1.1B-Chat-v1.0F161.1B2.05 GiB0.77 GiB3.71 GiB0.01 GiB32±26.5%
Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKINGI1-Q2_K4.5B1.67 GiB1.13 GiB3.71 GiB0.01 GiB32±26.5%
Qwen3.5-4B-SOMPOA-heresy-v2I1-Q2_K4.5B1.67 GiB1.13 GiB3.71 GiB0.01 GiB32±26.5%
Qwen3.5-4B-SOMPOA-heresyI1-Q2_K4.5B1.67 GiB1.13 GiB3.71 GiB0.01 GiB32±26.5%
Qwen3.5-4B-Safety-ThinkingI1-Q2_K4.2B1.67 GiB1.13 GiB3.71 GiB0.01 GiB32±26.5%
Huihui-Qwen3.5-4B-abliteratedI1-Q2_K4.5B1.67 GiB1.13 GiB3.71 GiB0.01 GiB32±26.5%
Darkidol-Ballad-4BI1-Q2_K4.5B1.67 GiB1.13 GiB3.71 GiB0.01 GiB32±26.5%
Qwen3.5-2B-hereticQ8_02.2B2.39 GiB0.42 GiB3.71 GiB0.01 GiB32±26.5%
dolphin-2_6-phi-2Q8_02.8B2.75 GiB0.00 GiB3.70 GiB0.02 GiB32±26.5%
EXAONE-4.0-1.2B-abliteratedI1-IQ3_M1.5B0.70 GiB2.11 GiB3.69 GiB0.03 GiB32±26.5%
Gemma-3-4b-it-Uncensored-DBL-XI1-IQ3_XS4.7B1.88 GiB0.90 GiB3.69 GiB0.03 GiB32±26.5%
GRM-Kerlin-3bI1-IQ3_M3.4B1.51 GiB1.27 GiB3.69 GiB0.03 GiB32±26.5%
Garnet-OCR-3B-0422I1-IQ3_M4.1B1.51 GiB1.27 GiB3.69 GiB0.03 GiB32±26.5%
Darwin-4B-ChimeraIQ3_XS4.0B1.78 GiB0.99 GiB3.68 GiB0.04 GiB32±26.5%
Vikhr-Gemma-2B-instructIQ2_XXS2.6B0.88 GiB1.89 GiB3.68 GiB0.04 GiB32±26.5%
Gemmasutra-Mini-2B-v1I1-IQ2_XXS2.6B0.88 GiB1.89 GiB3.68 GiB0.04 GiB32±26.5%
granite-4.0-h-microQ6_K_L3.2B2.49 GiB0.28 GiB3.67 GiB0.05 GiB32±26.5%
Holo-3.1-4BI1-IQ2_XXS5.2B1.63 GiB1.13 GiB3.66 GiB0.06 GiB32±26.5%
AfriqueQwen3.5-4BI1-IQ2_XXS5.2B1.63 GiB1.13 GiB3.66 GiB0.06 GiB32±26.5%
TimeOmni-1-4BI1-IQ2_XXS5.2B1.63 GiB1.13 GiB3.66 GiB0.06 GiB32±26.5%
Qwen3.5-4B-NSFW-ARA-Heretic-LiteroticaI1-IQ2_M4.2B1.62 GiB1.13 GiB3.66 GiB0.06 GiB33±26.5%
Qwen3.5-4B-RpRMax-v1I1-IQ2_M4.7B1.62 GiB1.13 GiB3.66 GiB0.06 GiB33±26.5%
Holo-3.1-4B-uncensored-hereticI1-IQ2_M4.5B1.62 GiB1.13 GiB3.66 GiB0.06 GiB33±26.5%
GRaPE-2-MiniI1-IQ2_M4.7B1.62 GiB1.13 GiB3.66 GiB0.06 GiB33±26.5%
Qwen3.5-DPO-4B-2I1-IQ2_M4.2B1.62 GiB1.13 GiB3.66 GiB0.06 GiB33±26.5%
Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliteratedI1-IQ2_M4.7B1.62 GiB1.13 GiB3.66 GiB0.06 GiB33±26.5%
Qwopus3.5-4B-v3-hereticI1-IQ2_M4.5B1.62 GiB1.13 GiB3.66 GiB0.06 GiB33±26.5%
Aureth-4B-Qwen3.5I1-IQ2_M4.5B1.62 GiB1.13 GiB3.66 GiB0.06 GiB33±26.5%
Dolphin3.0-Qwen2.5-3bQ3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
Qwen2.5-Coder-3B-Instruct-abliteratedI1-Q3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
GRM-Kerlin-3b-AbliteratedI1-Q3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
Mythos-nanoI1-Q3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
MATE-3BI1-Q3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
Mythos-nano-OBLITERATEDI1-Q3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
Qwen2.5-3B-Instruct-UncensoredI1-Q3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
Nanonets-OCR-sQ3_K_M3.8B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
Qwen2.5-Coder-3BQ3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
VibeThinker-3B-OBLITERATEDI1-Q3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
raspberry-3BQ3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
VibeThinker-3BQ3_K_M3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
Fourier-Qwen2.5-VL-3B-0.67I1-Q3_K_M3.8B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
Qwen2.5-VL-3B-InstructQ3_K_M3.8B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
jina-embeddings-v4Q3_K_M3.8B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
Qwen2.5-3BQ3_K_S3.1B1.48 GiB1.27 GiB3.66 GiB0.06 GiB33±26.5%
gemma-3n-E2B-itQ3_K_M5.4B2.31 GiB0.44 GiB3.65 GiB0.07 GiB33±26.5%
DeepScaleR-1.5B-PreviewQ8_01.8B1.76 GiB0.98 GiB3.65 GiB0.07 GiB32±26.5%
Qwen2.5-Coder-1.5B-Instruct-abliteratedQ8_01.8B1.76 GiB0.98 GiB3.65 GiB0.07 GiB32±26.5%
Qwen2.5-1.5B-Instruct-uncensoredQ8_01.8B1.76 GiB0.98 GiB3.65 GiB0.07 GiB32±26.5%
VibeThinker-1.5BQ8_01.8B1.76 GiB0.98 GiB3.65 GiB0.07 GiB32±26.5%
Nemotron-Research-Reasoning-Qwen-1.5BQ8_01.8B1.76 GiB0.98 GiB3.65 GiB0.07 GiB32±26.5%
DeepSeek-R1-Distill-Qwen-1.5B-uncensoredQ8_01.8B1.76 GiB0.98 GiB3.65 GiB0.07 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?
342 of 2118 indexed open-weight models fit a Radeon RX 6500 XT at 131,072 context with q4_0 KV cache, the largest being Trinity-Nano-Preview at Q2_K_L. 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.