Leaderboard Models Chips Compare People World Quickstart

Models

Loading community benchmarks…

Community throughput benchmarks for local AI.

Browse

Leaderboard
Models
Chips

Community

People
World

Contribute

Quickstart
ComputeArena CLI

Privacy Policy Terms of Service Contact

Operated by Base Compute Pty. Ltd.

Submissions524

Models37

Chips16

Contributors9

Submissions524

Models37

Chips16

Contributors9

Submissions524

Models37

Chips16

Contributors9

Submissions524

Models37

Chips16

Contributors9

Submissions524

Models37

Chips16

Contributors9

Submissions524

Models37

Chips16

Contributors9

Submissions524

Models37

Chips16

Contributors9

Submissions524

Models37

Chips16

Contributors9

Community benchmarks

Models

Explore the models our community is running. Open a model to compare its submitted benchmarks across chips, formats, and backends.

Add your benchmark

Models tested: 37
Submitted runs: 524

Search

Backend

Runtime

Quantisation

Prefill size

Sort by

37 of 37 models

Qwen3 8B

Peak decode · TG128: 150.7tok/s
Peak prefill · PP512: 3,241tok/s

Runs: 65
Chips: 9
Contributors: 3

BaseRTllama.cpp

Last run 20 Sep 2026

Qwen3 0.6B

Peak decode · TG128: 708.3tok/s
Peak prefill · PP512: 52,750tok/s

Runs: 35
Chips: 11
Contributors: 5

BaseRTllama.cpp

Last run 19 Sep 2026

Llama 3.1 8B Instruct

Peak decode · TG128: 136.1tok/s
Peak prefill · PP512: 7,567tok/s

Runs: 30
Chips: 9
Contributors: 2

BaseRTllama.cpp

Last run 20 Sep 2026

GPT OSS 20B

Peak decode · TG128: 195.6tok/s
Peak prefill · PP512: 4,424tok/s

Runs: 27
Chips: 10
Contributors: 3

BaseRTllama.cpp

Last run 20 Sep 2026

Gemma 4 E4B IT

Peak decode · TG128: 126.6tok/s
Peak prefill · PP512: 8,156tok/s

Runs: 26
Chips: 11
Contributors: 4

BaseRTllama.cpp

Last run 20 Sep 2026

Llama 3.2 3B Instruct

Peak decode · TG128: 229.7tok/s
Peak prefill · PP512: 14,988tok/s

Runs: 26
Chips: 8
Contributors: 2

BaseRTllama.cpp

Last run 20 Sep 2026

Qwen3.6 27B

Peak decode · TG128: 48.0tok/s
Peak prefill · PP512: 1,174tok/s

Runs: 24
Chips: 9
Contributors: 2

BaseRTllama.cpp

Last run 20 Sep 2026

Qwen3 30B A3B Instruct 2507

Peak decode · TG128: 198.1tok/s
Peak prefill · PP512: 7,439tok/s

Runs: 23
Chips: 7
Contributors: 2

BaseRTllama.cpp

Last run 20 Sep 2026

Qwen3 4B Instruct 2507

Peak decode · TG128: 185.1tok/s
Peak prefill · PP512: 11,246tok/s

Runs: 23
Chips: 8
Contributors: 2

BaseRTllama.cpp

Last run 20 Sep 2026

Qwen3.8 27B

Peak decode · TG128: 95.3tok/s
Peak prefill · PP512: 2,513tok/s

Runs: 20
Chips: 8
Contributors: 5

BaseRTllama.cpp

Last run 20 Sep 2026

Qwen3.6 35B A3B

Peak decode · TG128 @ 1 ctx: 139.7tok/s
Peak prefill · PP512: 2,597tok/s

Runs: 18
Chips: 8
Contributors: 3

BaseRTllama.cpp

Last run 20 Sep 2026

Gemma 3 1B IT

Peak decode · TG128: 415.5tok/s
Peak prefill · PP512: 33,841tok/s

Runs: 16
Chips: 7
Contributors: 3

Last run 20 Sep 2026

Nemotron 3 Nano 30B A3B

Peak decode · TG128: 184.4tok/s
Peak prefill · PP512: 4,984tok/s

Runs: 16
Chips: 6
Contributors: 3

Last run 20 Sep 2026

Gemma 4 E2B IT

Peak decode · TG128: 197.5tok/s
Peak prefill · PP512: 20,099tok/s

Runs: 15
Chips: 8
Contributors: 4

Last run 20 Sep 2026

Llama 3.2 1B Instruct

Peak decode · TG128 @ 1 ctx: 410.1tok/s
Peak prefill · PP512: 36,647tok/s

Runs: 15
Chips: 6
Contributors: 2

Last run 20 Sep 2026

Qwen3.5 2B

Peak decode · TG128 @ 1 ctx: 306.7tok/s
Peak prefill · PP512: 6,837tok/s

Runs: 14
Chips: 7
Contributors: 3

Last run 20 Sep 2026

Qwen3 1.7B

Peak decode · TG128 @ 1 ctx: 318.4tok/s
Peak prefill · PP512: 8,043tok/s

Runs: 13
Chips: 6
Contributors: 2

Last run 20 Sep 2026

Qwen3 4B

Peak decode · TG128 @ 1 ctx: 165.7tok/s
Peak prefill · PP512: 3,412tok/s

Runs: 13
Chips: 6
Contributors: 3

BaseRTllama.cpp

Last run 20 Sep 2026

Gemma 4 26B A4B IT

Peak decode · TG128: 92.0tok/s
Peak prefill · PP512: 6,190tok/s

Runs: 12
Chips: 6
Contributors: 3

Last run 20 Sep 2026

Mistral 7B Instruct v0.3

Peak decode · TG128 @ 1 ctx: 129.3tok/s
Peak prefill · PP512: 7,554tok/s

Runs: 12
Chips: 6
Contributors: 2

Last run 20 Sep 2026

Qwen3 4B Thinking 2507

Peak decode · TG128 @ 1 ctx: 155.6tok/s
Peak prefill · PP512: 11,473tok/s

Runs: 12
Chips: 6
Contributors: 2

Last run 20 Sep 2026

Qwen3.5 35B A3B

Peak decode · TG128 @ 1 ctx: 140.1tok/s
Peak prefill · PP512: 2,521tok/s

Runs: 12
Chips: 5
Contributors: 2

Last run 20 Sep 2026

Qwen3.5 2B Base

Peak decode · TG128 @ 1 ctx: 295.1tok/s
Peak prefill · PP512: 4,191tok/s

Runs: 11
Chips: 6
Contributors: 2

Last run 20 Sep 2026

Qwen3 30B A3B Thinking 2507

Peak decode · TG128 @ 1 ctx: 140.1tok/s
Peak prefill · PP512: 4,738tok/s

Runs: 10
Chips: 5
Contributors: 2

Last run 20 Sep 2026

Page 1 of 2

Community throughput benchmarks for local AI.

Browse

Leaderboard
Models
Chips

Community

People
World

Contribute

Quickstart
ComputeArena CLI

524 submissions · 37 models · 16 chips

Privacy Policy Terms of Service Contact

Operated by Base Compute Pty. Ltd.