Performance ranking (third-party evaluations)

Performance ranking of models that run in 24 GB of VRAM (LMArena)

Scores computed from people voting between two models' answers (Bradley-Terry, with style control). The range is the 95% interval; differences between models whose ranges overlap are not clear.

Only models with a quantization whose estimated memory fits in 24 GB at an 8,192-token context.

Top 29 of 29 models listed on this site

Values as of Nov 9, 2026

Rank Model Score (95% interval) Quantization that fits 24 GB
1

Qwen3.8-27B
Qwen 27.3B qwen3.8-27b

Score (95% interval) 1,433 1,429–1,438

32,659 votes

Q6_K 25.6 GB
2

Qwen3.5-27B
Qwen 26.9B qwen3.5-27b

Score (95% interval) 1,409 1,404–1,413

28,424 votes

Q6_K 25.4 GB
3

Qwen3-Next-80B-A3B-Instruct
Qwen 79.7B qwen3-next-80b-a3b-instruct

Score (95% interval) 1,399 1,394–1,403

22,692 votes

UD-IQ1_M 25.5 GB
4

Qwen3.5-35B-A3B
Qwen 34.7B qwen3.5-35b-a3b

Score (95% interval) 1,394 1,390–1,398

30,206 votes

Q4_1 24.1 GB
5

Qwen3-30B-A3B-Instruct-2507
Qwen 30.5B Qwen3-30B-A3B-Instruct (2507)

Score (95% interval) 1,382 1,377–1,387

23,287 votes

Q5_K_L 23.6 GB
6

Qwen3-Next-80B-A3B-Thinking
Qwen 79.7B qwen3-next-80b-a3b-thinking

Score (95% interval) 1,369 1,364–1,375

13,450 votes

UD-IQ1_M 25.4 GB
7

gemma-3-27b-it
google 27B Gemma 3 27B IT

Score (95% interval) 1,365 1,362–1,369

46,613 votes

Q6_K 24.6 GB
8

GLM-4.7-Flash
zai-org 29.9B glm-4.7-flash

Score (95% interval) 1,364 1,359–1,370

11,937 votes

Q5_K_L 23.1 GB
9

Qwen3-32B
Qwen 32.8B qwen3-32b

Score (95% interval) 1,347 1,337–1,356

3,926 votes

Q5_K_S 25.6 GB
10

gemma-3-12b-it
google 11.8B Gemma 3 12B IT

Score (95% interval) 1,342 1,332–1,351

3,829 votes

BF16 25.7 GB
11

granite-4.2-30b
ibm-granite 29.3B Granite 4.2 30B

Score (95% interval) 1,340 1,329–1,350

3,314 votes

Q5_1 24.9 GB
12

QwQ-32B
Qwen 32.8B qwq-32b

Score (95% interval) 1,336 1,331–1,340

25,166 votes

Q5_K_S 25.6 GB
13

Qwen3-30B-A3B
Qwen 30.5B qwen3-30b-a3b

Score (95% interval) 1,326 1,322–1,331

26,137 votes

UD-Q5_K_XL 23.4 GB
14

Llama-3.3-70B-Instruct
meta-llama 70.6B Llama 3.3 70B Instruct

Score (95% interval) 1,318 1,315–1,322

54,482 votes

IQ2_S 25.7 GB
15

gpt-oss-20b
openai 20.9B gpt-oss-20b

Score (95% interval) 1,318 1,311–1,324

10,400 votes

F16 15.0 GB
16

gemma-3n-E4B-it
google 6.9B Gemma 3n E4B IT

Score (95% interval) 1,317 1,312–1,323

22,106 votes

F16 14.9 GB
17

gemma-3-4b-it
google 3.9B Gemma 3 4B IT

Score (95% interval) 1,303 1,294–1,313

4,171 votes

BF16 9.19 GB
18

gemma-2-27b-it
google 27.2B Gemma 2 27B IT

Score (95% interval) 1,290 1,286–1,293

75,754 votes

Q5_K_L 23.2 GB
19

granite-4.2-8b
ibm-granite 8.8B Granite 4.2 8B

Score (95% interval) 1,290 1,278–1,301

3,196 votes

BF16 19.7 GB
20

granite-4.2-3b
ibm-granite 3.7B Granite 4.2 3B

Score (95% interval) 1,289 1,277–1,301

3,185 votes

BF16 8.74 GB
21

Mistral-Small-24B-Instruct-2501
mistralai 23.6B Mistral Small 24B Instruct (2501)

Score (95% interval) 1,274 1,268–1,280

14,681 votes

Q6_K 21.5 GB
22

Qwen2.5-Coder-32B-Instruct
Qwen 32.8B qwen2.5-coder-32b-instruct

Score (95% interval) 1,271 1,263–1,279

5,432 votes

Q5_K_S 25.6 GB
23

gemma-2-9b-it
google 9.2B Gemma 2 9B IT

Score (95% interval) 1,267 1,263–1,271

54,611 votes

Q8_0 14.0 GB
24

phi-4
microsoft 14.7B phi-4

Score (95% interval) 1,256 1,252–1,261

24,126 votes

Q8_0 18.0 GB
25

gemma-2-2b-it
google 2.6B Gemma 2 2B IT

Score (95% interval) 1,200 1,196–1,204

46,616 votes

F32 12.2 GB
26

Llama-3.2-3B-Instruct
meta-llama 3.2B Llama 3.2 3B Instruct

Score (95% interval) 1,167 1,159–1,175

7,936 votes

F16 8.17 GB
27

Mistral-7B-Instruct-v0.2
mistralai 7.2B Mistral 7B Instruct v0.2

Score (95% interval) 1,149 1,143–1,156

19,402 votes

Q8_0 9.37 GB
28

Llama-3.2-1B-Instruct
meta-llama 1.2B Llama 3.2 1B Instruct

Score (95% interval) 1,111 1,103–1,119

8,045 votes

F16 3.55 GB
29

gemma-2b-it
lmstudio-ai 2.5B Gemma 2B IT

Score (95% interval) 1,094 1,082–1,105

4,780 votes

Q8_0 3.89 GB

Estimated Memory at 8K: Estimated with a formula. Actual usage depends on your setup and settings.

About this ranking

Values for open-weight models come from the latest version of LMArena's Hugging Face dataset (text, style control, overall). Updated weekly.
Values are for the original model before quantization. GGUF quantization (especially small quantizations) can lower performance.
Values are attached only when the model name in the source matches a model listed on this site (its original Hugging Face repository). Fine-tunes and modified versions do not get them.
Each source is a separate ranking. We do not mix values from different sources or assign our own scores.
Small and new models are often not evaluated by the source, so they do not appear here.

Values are from LMArena data (CC BY 4.0), shown unchanged (sorting and filtering by this site).