Dataset Viewer
Auto-converted to Parquet Duplicate
vram_gb
int64
7
128
hardware_kind
stringclasses
3 values
hardware
stringlengths
12
55
model
stringlengths
10
61
run_count
int64
3
48
avg_quality
float64
10.3
87.2
avg_tok_s
float64
15.2
172
scenario_coding
float64
0
83.3
scenario_agent
float64
28
93
scenario_roleplay
float64
7.7
93.3
scenario_research
float64
5.6
88.5
scenario_coding_runs
int64
3
48
scenario_agent_runs
int64
3
48
scenario_roleplay_runs
int64
3
48
scenario_research_runs
int64
3
48
last_updated
stringdate
2026-08-13 00:00:00
2026-09-10 00:00:00
source
stringclasses
1 value
128
Apple Silicon
Apple M5 Max
mtplx-qwen38-27b-optimized-quality
3
87.2
35
78.4
90.6
92.4
87.6
3
3
3
3
2026-09-01
https://llm-bench.io
128
Apple Silicon
Apple M5 Max
Qwen3.8-27B-oQ4e-fp16-mtp
4
85.1
38.6
82.5
85.1
89.6
83
4
4
4
4
2026-08-20
https://llm-bench.io
128
Apple Silicon
Apple M5 Max
mtplx-flash-next-bare-speed
3
85
57.5
75.6
84.6
93.3
86.3
3
3
3
3
2026-09-01
https://llm-bench.io
128
Apple Silicon
Apple M4 Max
Qwen3.8-Flash-Next-oQ5e-mtp
10
84
37.6
76
84.7
89.8
85.5
10
10
10
10
2026-09-09
https://llm-bench.io
128
Apple Silicon
Apple M4 Max
Qwen3.8-Flash-Next-oQ4e-mtp
17
82.6
45.3
76.5
82.5
86.8
84.7
17
17
17
17
2026-09-07
https://llm-bench.io
128
Apple Silicon
Apple M2 Ultra
Qwen3.8-Flash-Next-oQ4e-mtp
5
80.9
25.2
59.3
87
91.6
85.7
5
5
5
5
2026-09-01
https://llm-bench.io
128
Apple Silicon
Apple M2 Ultra
Qwen3.8-27B-oQ8e-fp16-mtp
13
80.7
34.1
72.7
77.3
88.2
84.6
13
13
13
13
2026-09-04
https://llm-bench.io
128
Apple Silicon
Apple M4 Max
Ling-3.0-tiny-oQ8e
3
58.7
126
45.2
68.2
43.6
77.5
3
3
3
3
2026-09-10
https://llm-bench.io
96
NVIDIA
NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition
Qwen3.8-27B-UD-Q6_K_XL
4
79.5
84.3
75.9
76.9
84.4
80.6
4
4
4
4
2026-08-31
https://llm-bench.io
64
Apple Silicon
Apple M4 Max
mtplx-qwen38-27b-optimized-quality
3
85.7
35.5
77.7
85.2
93.1
86.7
3
3
3
3
2026-09-02
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Qwen3.8-27B-oQ8e-fp16-mtp
4
84.6
33.7
74.1
87.4
91.2
85.6
4
4
4
4
2026-08-28
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Muse-Glimmer-30B-oQ8e
3
83.9
17
72.3
89.4
92
82.1
3
3
3
3
2026-08-29
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Qwen3.8-27B-oQ4-mtp
4
83.9
42.8
75.4
89.2
85
86.1
4
4
4
4
2026-08-28
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Ornith-1.5-35B-A3B-oQ8e-mtp
5
83.7
94.6
77.5
82
88.5
86.8
5
5
5
5
2026-09-07
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Qwen3.8-27B-oQ4e-mtp
26
83.5
45.3
74.2
83.4
90.5
86
26
26
26
26
2026-09-09
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Tiel-Coder-35B-A3B-MLX-oQ4e
4
83.4
118.4
80.7
82.7
85.6
84.4
4
4
4
4
2026-08-25
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Tiel-Coder-35B-A3B-MLX-oQ4e-MTP
5
82.1
123.7
76.3
84.4
82.3
85.3
5
5
5
5
2026-09-07
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Muse-Glimmer-30B-4bit
3
82.1
29.9
68.6
89.3
89.6
80.9
3
3
3
3
2026-08-13
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Qwen3.8-27B-4bit
9
81.9
30.6
76.8
78
89.2
83.4
9
9
9
9
2026-08-22
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Qwen3.8-27B-oQ8e-mtp
26
81.3
33.4
70.7
82.2
88.1
84.4
26
26
26
26
2026-09-10
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
unsloth/Qwen3.8-27B-GGUF:UD-Q8_K_XL
3
80.9
24.4
68.1
80.7
89.3
85.4
3
3
3
3
2026-08-21
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Qwen3.8-27B-oQ8-mtp
4
77.7
35.6
77.6
67.4
84.8
81
4
4
4
4
2026-08-17
https://llm-bench.io
64
Apple Silicon
Apple M5 Max
Qwen3.8-27B-oQ2e-mtp
3
10.3
44
0
28
7.7
5.6
3
3
3
3
2026-08-30
https://llm-bench.io
23
NVIDIA
NVIDIA GeForce RTX 4090
unsloth/Qwen3.8-27B-GGUF:IQ3_S
4
86.9
108.6
83.3
86.2
90.3
87.9
4
4
4
4
2026-08-29
https://llm-bench.io
23
NVIDIA
NVIDIA GeForce RTX 4090
unsloth/Qwen3.8-27B-GGUF:Q4_K_M
6
85.1
95.3
78
85.9
89
87.4
6
6
6
6
2026-08-29
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XT/7900 XTX/7900 GRE/7900M
qwen3.8-27b-UD-Q4_K_XL
3
84.9
50.7
80.6
83.7
86.9
88.5
3
3
3
3
2026-09-08
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XT/7900 XTX/7900 GRE/7900M
Tiel-Coder-35B-A3B-Q4_K_S
3
84.6
165
76.9
89.3
87.3
84.9
3
3
3
3
2026-09-05
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XTX
qwen3.8:27b-mtp-q4_K_M
3
83.5
40.8
71.2
88.2
88.1
86.7
3
3
3
3
2026-08-16
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XTX
qwen3.8:latest
7
82.6
37.4
75.9
82.2
85.1
87.1
7
7
7
7
2026-09-07
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XTX
qwen3.8:27b
4
82.2
38.4
77.6
81.1
85
84.8
4
4
4
4
2026-08-18
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XTX
muse-glimmer:latest
4
82
34.5
71.5
93
79.7
83.6
4
4
4
4
2026-08-25
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XT/7900 XTX/7900 GRE/7900M
Qwen3.8-27B-Q4_K_XL
15
81.5
72.4
78.7
76.5
85.7
85
15
15
15
15
2026-09-05
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XT/7900 XTX/7900 GRE/7900M
Ornith-1.5-35B-A3B-IQ4_XS
3
80.7
139.5
78.7
81.4
82.4
80.2
3
3
3
3
2026-09-05
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XT/7900 XTX/7900 GRE/7900M
qwen38-27b
6
79.6
72.7
67.5
77.3
88.7
85
6
6
6
6
2026-09-08
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XT/7900 XTX/7900 GRE/7900M
Qwen3.6-35B-A3B-IQ4_NL
17
75.9
164.8
67.1
74.7
83
78.7
17
17
17
17
2026-09-05
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XT/7900 XTX/7900 GRE/7900M
gemma-4-26b-a4b-q4kxl
3
69.5
169.9
67.9
75.2
79.9
55.2
3
3
3
3
2026-09-05
https://llm-bench.io
23
AMD
AMD Radeon RX 7900 XTX
gemma4:26b-a4b-it-qat
3
69.2
116
60.5
81.7
70.1
64.3
3
3
3
3
2026-08-24
https://llm-bench.io
20
NVIDIA
NVIDIA GeForce RTX 3080 Ti
unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF:UD-Q4_K_XL
3
79.7
172.4
68
84.5
83.2
83.1
3
3
3
3
2026-09-07
https://llm-bench.io
16
NVIDIA
NVIDIA GeForce RTX 4080 SUPER
Qwen3.8-27B-GSQ-RCO-IQ3_S
4
83.8
73.6
75.3
84.7
88.8
86.3
4
4
4
4
2026-09-04
https://llm-bench.io
16
NVIDIA
NVIDIA GeForce RTX 4080 SUPER
Qwen3.8-27B-UD-IQ3_S
48
83.2
78.8
77.8
81.6
87.8
85.7
48
48
48
48
2026-09-05
https://llm-bench.io
16
AMD
Advanced Micro Devices, Inc. [AMD/ATI] HawkPoint1
peculiar-ragdoll/Tiel-Coder-35B-A3B-GGUF-MTP
4
78.9
26.8
77.2
81.8
80.9
75.8
4
4
4
4
2026-09-01
https://llm-bench.io
16
AMD
Advanced Micro Devices, Inc. [AMD/ATI] HawkPoint1
mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF
4
66.8
31.6
65.7
59.9
68
73.8
4
4
4
4
2026-09-01
https://llm-bench.io
15
NVIDIA
NVIDIA GeForce RTX 5070 Ti
Qwen3.8-27B-UD-Q3_K_XL
8
84.6
66.3
73.3
86.8
92.3
86.2
8
8
8
8
2026-09-04
https://llm-bench.io
10
NVIDIA
NVIDIA GeForce RTX 3080
qwen3_coder_next
4
80.1
31.2
72.5
87.3
85.3
75.4
4
4
4
4
2026-08-25
https://llm-bench.io
10
NVIDIA
NVIDIA GeForce RTX 3080
qwen36-35b_VISION
5
67
33.1
60.7
52.9
81.6
73
5
5
5
5
2026-08-26
https://llm-bench.io
10
NVIDIA
NVIDIA GeForce RTX 3080
qwen35_122b
5
66.7
15.2
53.7
56.3
80.3
76.3
5
5
5
5
2026-08-25
https://llm-bench.io
10
NVIDIA
NVIDIA GeForce RTX 3080
ornith-1.5-9b
3
54.3
93.8
11.7
55.2
76.6
73.8
3
3
3
3
2026-08-25
https://llm-bench.io
8
NVIDIA
NVIDIA GeForce RTX 5060
Tiel-Coder-35B-A3B-MTP-UD-IQ3_XXS
3
83.7
90
78.4
86.7
85.4
84.3
3
3
3
3
2026-09-09
https://llm-bench.io
8
NVIDIA
NVIDIA GeForce RTX 5060
Cyber-Tiel-Coder-35B-A3B-MTP-UD-IQ3_XXS
3
81.5
115.7
73.8
85.3
82.4
84.7
3
3
3
3
2026-09-10
https://llm-bench.io
8
NVIDIA
NVIDIA GeForce RTX 5060
gpt-oss-20b-UD-Q8_K_XL
3
75.6
148.4
70.3
83.8
71.6
76.7
3
3
3
3
2026-09-09
https://llm-bench.io
7
AMD
AMD Radeon RX 5600 OEM/5600 XT / 5700/5700 XT
Tiel-Coder-35B-A3B-MTP-UD-Q6_K_XL
3
85.5
35.1
83.3
90.5
82.6
85.6
3
3
3
3
2026-09-04
https://llm-bench.io

YAML Metadata Warning:The task_categories "tabular-data" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other

llm-bench.io — Community LLM Benchmark Leaderboard by Hardware

Per-model community benchmark data for local LLMs, curated from llm-bench.io and grouped by hardware and available VRAM.

This dataset contains only aggregated statistics derived from individual benchmark submissions. It does not contain raw submissions, prompts, model responses, machine identifiers, or client/session data. The raw data lives in the llm-bench.io database and is used here only to compute the summary stats below.

Why this dataset

LLM enthusiasts want to know: "How well does this model run on my hardware, and how good is the output?" This table gives a trustworthy, community-sized answer by filtering to model/hardware combinations with enough independent runs to be meaningful.

What's in it

Column Description
vram_gb Available VRAM (or unified memory) in GB on the tested hardware
hardware_kind Platform category: Apple Silicon, NVIDIA, AMD
hardware Specific device name (e.g. Apple M5 Max, NVIDIA GeForce RTX 4090)
model Model name as run by the community (e.g. Qwen3.8-27B-oQ4e-mtp)
run_count Number of independent community runs aggregated for this model/hardware
avg_quality Mean quality score (0–100) across all four scenarios
avg_tok_s Average generation speed in tokens/second
scenario_coding Mean quality score for the code generation scenario (0–100)
scenario_agent Mean quality score for long-horizon agent / task scenarios (0–100)
scenario_roleplay Mean quality score for role-play / narrative scenarios (0–100)
scenario_research Mean quality score for research & analysis scenarios (0–100)
scenario_*_runs Number of runs feeding each scenario score
last_updated ISO date of the most recent run contributing to this row
source Link to the live site where results can be explored interactively

Methodology

Quality scores. Each submission includes scores on four tasks (code generation, long-horizon agent workflows, role-play, and research) that are scored by an automated LLM judge (a model used as an evaluator, not a human rater). Scores are 0–100.

Aggregation. For every (model, hardware) pair we:

  1. Take all runs from the last 30 days.
  2. Require at least 3 independent runs (this filters out noise and outliers — a single run is not trusted).
  3. Average the per-scenario scores across those runs.

Hardware grouping. Rows are grouped by the device a model was tested on, so a 32 tok/s figure on a laptop and a 32 tok/s figure on a desktop are never mixed.

Privacy. The published table is a statistical summary. Individual submissions, prompts, model outputs, machine hashes, and client/session identifiers are never exported — they remain only in the llm-bench.io database.

How to use

import pandas as pd

df = pd.read_csv("benchmarks-by-vram.csv")

# Best coding model on an RTX 4090 with at least 4 runs
df[(df.hardware.str.contains("4090")) & (df.run_count >= 4)] \
    .sort_values("scenario_coding", ascending=False).head()

Or load directly in Python:

from datasets import load_dataset

ds = load_dataset("llmbenchio/benchmarks-by-vram")

License

Data is community-sourced and made available for reference and research use. Attribution is appreciated: source is llm-bench.io.

Downloads last month
39

Space using llmbenchio/benchmarks-by-vram 1