fedi-index: rangering af AI-modeller på åbne data

GPU 12 GB

1. 1 Qwen3 Next 80B A3B Instruct Q5_K_M · delvist i RAM ≈24 tok/s 62
2. 2 Gemma 4 26B A4B UD-Q8_K_XL · delvist i RAM ≈15 tok/s 54
3. 3 Qwen3.5-35B-A3B Q8_0 · delvist i RAM ≈18 tok/s 52
4. 4 Qwen3.6 35B A3B UD-Q8_K_XL · delvist i RAM ≈17 tok/s 47
5. 5 Qwen3 Next 80B A3B Thinking Q5_K_M · delvist i RAM ≈24 tok/s 46
6. 6 Qwen3.5-122B-A10B UD-IQ4_NL · delvist i RAM ≈10 tok/s 45
7. 7 Qwen3 30B A3B Instruct 2507 Q8_0 · delvist i RAM ≈16 tok/s 45
8. 8 gpt-oss-120b MXFP4 · delvist i RAM ≈19 tok/s 44
9. 9 Qwen3.5-9B Q6_K ≈36 tok/s 42
10. 10 Qwen3.8 27B UD-Q4_K_XL · delvist i RAM ≈4 tok/s 37

GPU 24 GB

1. 1 Qwen3.8 27B UD-Q5_K_S ≈38 tok/s 81
2. 2 Gemma 4 31B Q4_1 ≈37 tok/s 67
3. 3 Gemma 4 26B A4B UD-Q5_K_S ≈129 tok/s 63
4. 4 Qwen3 Next 80B A3B Instruct UD-Q6_K_XL · delvist i RAM ≈26 tok/s 62
5. 5 Qwen3.5-27B Q5_K_S ≈38 tok/s 60
6. 6 Qwen3.6 27B Q5_K_S ≈37 tok/s 59
7. 7 Qwen3.5-35B-A3B Q4_K_S ≈191 tok/s 56
8. 8 Qwen3.6 35B A3B UD-Q4_K_S ≈189 tok/s 51
9. 9 Qwen3 30B A3B Instruct 2507 Q5_K_S ≈147 tok/s 50
10. 10 Qwen3.5-122B-A10B Q4_K_S · delvist i RAM ≈11 tok/s 47

Mac 64 GB

1. 1 Qwen3.8 27B Q8_0 ≈13 tok/s 64
2. 2 Gemma 4 26B A4B UD-Q8_K_XL ≈40 tok/s 63
3. 3 Qwen3 Next 80B A3B Instruct Q4_K_S ≈84 tok/s 62
4. 4 Qwen3.5-122B-A10B UD-IQ3_XXS ≈44 tok/s 56
5. 5 Qwen3.5-35B-A3B Q8_0 ≈51 tok/s 56
6. 6 Qwen3.6 35B A3B UD-Q8_K_XL ≈49 tok/s 51
7. 7 Gemma 4 31B Q8_0 ≈11 tok/s 50
8. 8 Qwen3 30B A3B Instruct 2507 Q8_0 ≈45 tok/s 50
9. 9 Qwen3.5-27B Q8_0 ≈13 tok/s 48
10. 10 Qwen3.6 27B Q8_0 ≈13 tok/s 47

128 GB delt hukommelse

1. 1 DeepSeek V4 Flash 0423 UD-IQ3_XXS ≈26 tok/s 65
2. 2 Inkling Small UD-IQ3_S ≈23 tok/s 65
3. 3 MiniMax M2.7 UD-IQ4_NL ≈20 tok/s 64
4. 4 Gemma 4 26B A4B UD-Q8_K_XL ≈29 tok/s 63
5. 5 Qwen3 Next 80B A3B Instruct Q8_0 ≈36 tok/s 62
6. 6 Qwen3.8 27B UD-Q4_K_M ≈10 tok/s 58
7. 7 Qwen3.5-35B-A3B Q8_0 ≈37 tok/s 56
8. 8 Step 3.5 Flash IQ4_XS ≈17 tok/s 55
9. 9 Qwen3.5-122B-A10B Q6_K ≈15 tok/s 54
10. 10 DeepSeek V4 Flash 0731 UD-IQ3_XXS ≈11 tok/s 53

CPU, 32 GB RAM

1. 1 Gemma 4 26B A4B UD-Q4_K_XL ≈6 tok/s 36
2. 2 Qwen3.5-35B-A3B Q4_K_M ≈7 tok/s 32
3. 3 Qwen3.6 35B A3B UD-Q4_K_XL ≈7 tok/s 30
4. 4 Qwen3 30B A3B Instruct 2507 Q4_K_M ≈6 tok/s 28
5. 5 Qwen3.8 27B UD-Q4_K_XL ≈2 tok/s 27
6. 6 GLM 4.7 Flash Q4_K_M ≈6 tok/s 24
7. 7 Nemotron 3 Nano 30B A3B Q4_K_M ≈7 tok/s 23
8. 8 Gemma 4 31B Q4_K_M ≈2 tok/s 22
9. 9 Qwen3 30B A3B Q4_K_M ≈6 tok/s 21
10. 10 Qwen3.5-27B Q4_K_M ≈2 tok/s 21