AI models on a 48 GB graphics card
Open chat models judged for a PC card with 48 GB of video memory: which fit with every part on the card, the size to download, and what each one takes.
3702 of the 3940 chat models people download as GGUF fit a 48 GB card whole. The best-rated is Gemma 4 31B at Q8_0, which takes 40 GB. The largest rated model that fits is Qwen3 Next 80B A3B (80B, 3.9B active).
Cards with 48 GB: RTX A6000, RTX 6000 Ada. The same answer holds for any card with 48 GB of its own memory.
The best-rated models that fit
The 25 with the highest LM Arena rating (data, CC BY 4.0) that fit at 4 bits a weight or more, each at the most bits that fits. Below 4 bits a model's rating overstates it. Each model's page hands it to LM Studio, Jan, Ollama and llama.cpp.
| Model | Size | Fit | Takes | Longest chat | Arena | Arena coding |
|---|---|---|---|---|---|---|
| Gemma 4 31BGoogle · 31B | Q8_033 GB · unsloth | Fits with room | 40 GB | 128K | 1452 | 1502 |
| Gemma 4 26BGoogle · 25B, 3.8B active | UD-Q6_K23 GB · unsloth | Fits with room | 27 GB | 256K | 1438 | 1482 |
| Qwen3.8 27BAlibaba Qwen · 27B | Q8_029 GB · unsloth | Fits with room | 40 GB | 256K | 1436 | 1498 |
| Qwen3.5 27BAlibaba Qwen · 27B | UD-Q8_K_XL36 GB · unsloth | Fits with room | 41 GB | 128K | 1409 | 1451 |
| Qwen3 Next 80B A3BAlibaba Qwen · 80B, 3.9B active | UD-Q4_K_XL46 GB · unsloth | Fits | 47 GB | 64K | 1399 | 1444 |
| Qwen3.5 35B A3BAlibaba Qwen · 36B, 3.5B active | Q8_037 GB · unsloth | Fits with room | 40 GB | 256K | 1394 | 1435 |
| Qwen3 Next 80B A3B ThinkingAlibaba Qwen · 80B, 3.9B active | UD-Q4_K_XL46 GB · unsloth | Fits | 47 GB | 64K | 1369 | 1421 |
| GLM 4.7 FlashZ.ai · 30B, 3.9B active | Q8_032 GB · unsloth | Fits with room | 36 GB | 128K | 1365 | 1423 |
| Gemma 3 27BGoogle · 27B | Q8_029 GB · unsloth | Fits with room | 35 GB | 128K | 1365 | 1357 |
| Qwen3 32BAlibaba Qwen · 33B | Q8_035 GB · Qwen | Fits with room | 40 GB | 32K | 1347 | 1406 |
| Llama 3 3 Nemotron Super 49B V1 5NVIDIA · 50B | Q5_K_M35 GB · unsloth | Fits with room | 39 GB | 32K | 1343 | 1403 |
| Gemma 3 12BGoogle · 12B | Q8_013 GB · lm-kit | Fits with room | 18 GB | 128K | 1342 | 1316 |
| Granite 4.2 30BIBM · 29B | Q8_031 GB · ibm-granite | Fits with room | 41 GB | 64K | 1340 | 1399 |
| QwQ 32BAlibaba Qwen · 33B | Q8_035 GB · Qwen | Fits with room | 40 GB | 32K | 1336 | 1383 |
| Olmo 3.1 32BAi2 · 32B | Q8_034 GB · unsloth | Fits with room | 37 GB | 64K | 1329 | 1382 |
| Llama 3 3 Nemotron Super 49B V1NVIDIA · 50B | Q5_K_M35 GB · bartowski | Fits with room | 39 GB | 32K | 1328 | |
| Qwen3 30B A3BAlibaba Qwen · 31B, 3.4B active | Q8_032 GB · Qwen | Fits with room | 34 GB | 32K | 1327 | 1385 |
| Llama 3.3 70BMeta · 71B | Q3_K_L37 GB · bartowski | Fits with room | 40 GB | 16K | 1318 | 1346 |
| Gemma 3n E4BGoogle · 6.9B | Q8_07.4 GB · unsloth | Fits with room | 7.7 GB | 32K | 1318 | 1307 |
| GPT-OSS 20BOpenAI · 21B, 4.2B active | UD-Q4_K_XL12 GB · unsloth | Fits with room | 14 GB | 128K | 1317 | 1368 |
| Olmo 3 32B ThinkAi2 · 32B | Q8_034 GB · unsloth | Fits with room | 37 GB | 64K | 1306 | 1360 |
| Granite 4.1 8BIBM · 8.8B | Q8_09.3 GB · ibm-granite | Fits with room | 22 GB | 128K | 1304 | 1353 |
| Gemma 3 4BGoogle · 3.9B | Q6_K_L3.4 GB · bartowski | Fits with room | 5.1 GB | 128K | 1303 | 1273 |
| Qwen2.5 72BAlibaba Qwen · 73B | Q3_K_L40 GB · ThomasBaruzier | Fits with room | 41 GB | 8K | 1303 | 1355 |
| Llama 3.1 Nemotron 70BNVIDIA · 71B | IQ4_XS38 GB · bartowski | Fits with room | 41 GB | 16K | 1299 | 1328 |
The chat models Outmake picks
18 of these 28 fit. Their memory is what llama.cpp's own calculator reported for each size, not an estimate.
| Model | Size | Fit | Takes |
|---|---|---|---|
| Qwen3.8 27BAlibaba Qwen · 28B | UD-Q6_K23 GB | Fits with room | 32 GB |
| Gemma 4 31BGoogle · 31B | Q6_K26 GB | Fits with room | 34 GB |
| K2-Horizon 36BInstitute of Foundation Models · 37B, 4B active | Q6_K32 GB | Fits with room | 36 GB |
| Gemma 4 26BGoogle · 26B, 3.8B active | Q8_028 GB | Fits with room | 30 GB |
| K2-Horizon 7BInstitute of Foundation Models · 9B | Q6_K7.8 GB | Fits with room | 29 GB |
| K2-Horizon 3.7BInstitute of Foundation Models · 5.1B | Q6_K4.4 GB | Fits with room | 26 GB |
| Qwen3.6 35BAlibaba Qwen · 36B, 3B active | UD-Q6_K30 GB | Fits with room | 32 GB |
| Gemma 4 12BGoogle · 12B | Q6_K10.0 GB | Fits with room | 13 GB |
| GLM 4.7 FlashZ.ai · 31B, 3B active | Q6_K25 GB | Fits with room | 29 GB |
| Qwen3.5 9BAlibaba Qwen · 9B | Q6_K8.4 GB | Fits with room | 14 GB |
| Gemma 4 E4BGoogle · 8B | UD-Q5_K_XL7.6 GB | Fits with room | 7.6 GB |
| Devstral Small 2 24BMistral AI · 24B | Q6_K20 GB | Fits with room | 32 GB |
| Gemma 4 E2BGoogle · 5.1B | Q6_K5.5 GB | Fits with room | 5.5 GB |
| Qwen3.5 4BAlibaba Qwen · 4B | Q6_K4.2 GB | Fits with room | 7.3 GB |
| GPT-OSS 20BOpenAI · 21B, 3.6B active | UD-Q8_K_XL13 GB | Fits with room | 14 GB |
| Qwen3.5 2BAlibaba Qwen · 2B | Q6_K2.2 GB | Fits with room | 3.6 GB |
| Qwen3.5 0.8BAlibaba Qwen · 0.8B | Q6_K0.8 GB | Fits with room | 2.4 GB |
| Qwen3.5 122BAlibaba Qwen · 122B, 10B active | UD-IQ3_XXS46 GB | Fits | 47 GB |
| GPT-OSS 120BOpenAI · 117B, 5.1B active | UD-Q4_K_XL63 GB | Bigger than the card | 65 GB |
| Qwen3.8 Flash NextAlibaba Qwen · 177B, 6B active | UD-Q3_K_XL91 GB | Bigger than the card | 91 GB |
| MiniMax M2.7MiniMax · 229B, 10B active | UD-Q3_K_XL102 GB | Bigger than the card | 120 GB |
| Qwen3.5 397BAlibaba Qwen · 403B, 17B active | UD-IQ2_M124 GB | Bigger than the card | 129 GB |
| DeepSeek V4 FlashDeepSeek · 284B, 13B active | UD-Q3_K_XL128 GB | Bigger than the card | 129 GB |
| GLM 5.3Z.AI · 754B, 40B active | UD-Q2_K_XL254 GB | Bigger than the card | 265 GB |
| Kimi K2.6Moonshot AI · 1T, 32B active | UD-Q2_K_XL341 GB | Bigger than the card | 351 GB |
| Kimi K3Moonshot AI · 2.8T, 104B active | UD-IQ2_XXS712 GB | Bigger than the card | 716 GB |
| Qwen3.8 2.4TAlibaba Qwen · 2.4T, 95B active | UD-IQ2_XS731 GB | Bigger than the card | 736 GB |
| DeepSeek V4 ProDeepSeek · 1.6T, 49B active | UD-Q4_K_XL850 GB | Bigger than the card | 851 GB |
How this is worked out
A model fits when its weights, its memory of the chat so far and the engine's own share come to 44 GB or less: 92% of the card, with the rest left for the screen. "Fits with room" is under 80%. "Tight fit" is anything up to half a gigabyte short of the whole card: close other programs that draw with the card first.
"Bigger than the card" does not mean it won't run. llama.cpp and the apps built on it put what doesn't fit in ordinary memory and the model answers slower, often several times slower. So this page lists what fits whole.
There are no speeds here. Outmake's speed estimates are fitted to measurements on Apple chips, and no PC has been measured yet, so a number would be a guess. Memory was checked against the catalog of November 4, 2026; which cards have 48 GB against each maker's specification page on October 13, 2026.