Original benchmark dataset

Local LLM benchmarks

Tokens-per-second rows collected on owner hardware or from cited, reproduced sources. Every public row ships with a confidence badge so you know which numbers are owner-measured, independently reproduced, or source-backed.

MMeasuredM~Measured-nearCSource-backed~ExtrapolatedEEstimatedHow we measure →
Browse all results ->Submit a result ->Read methodology ->See benchmark roadmap →Mobile + edge gap report →
Confidence engine
0 strong
Coverage map
0 wanted
Regression watch
version-aware
Reproduce
protocol + queue
Selected coverage slice

0 measured or reviewed / 0 estimated / 0 wanted / 168 unstarted

0.0% of 168 selected cells measured or reviewed

0 measured or reviewed cells, 0 estimated cells, 0 wanted cells, and 168 unstarted cells.

Selected benchmark coverage by hardware and model. Measured, reviewed, estimated, and wanted cells are visually distinct.
Hardware / Model
all-MiniLM-L6-v2
FLUX.1 [dev]
DeepSeek V4 Pro (1
Qwen 3 235B-A22B
Nomic Embed Text v
Kokoro 82M
Llama 4 Scout
Llama 3.1 8B Instr
DeepSeek V4 Flash
DeepSeek R1 (671B
BGE Large EN v1.5
Qwen 3 0.6B
Qwen 3 30B-A3B
Llama 3.3 70B Inst
NVIDIA DGX Spark (Project Di
all-MiniLM-L6-v2 on NVIDIA DGX Spark (Project Digits): not started
FLUX.1 [dev] on NVIDIA DGX Spark (Project Digits): not started
DeepSeek V4 Pro (1.6T MoE) on NVIDIA DGX Spark (Project Digits): not started
Qwen 3 235B-A22B on NVIDIA DGX Spark (Project Digits): not started
Nomic Embed Text v1.5 on NVIDIA DGX Spark (Project Digits): not started
Kokoro 82M on NVIDIA DGX Spark (Project Digits): not started
Llama 4 Scout on NVIDIA DGX Spark (Project Digits): not started
Llama 3.1 8B Instruct on NVIDIA DGX Spark (Project Digits): not started
DeepSeek V4 Flash (284B MoE) on NVIDIA DGX Spark (Project Digits): not started
DeepSeek R1 (671B reasoning) on NVIDIA DGX Spark (Project Digits): not started
BGE Large EN v1.5 on NVIDIA DGX Spark (Project Digits): not started
Qwen 3 0.6B on NVIDIA DGX Spark (Project Digits): not started
Qwen 3 30B-A3B on NVIDIA DGX Spark (Project Digits): not started
Llama 3.3 70B Instruct on NVIDIA DGX Spark (Project Digits): not started
Apple M1 Ultra
all-MiniLM-L6-v2 on Apple M1 Ultra: not started
FLUX.1 [dev] on Apple M1 Ultra: not started
DeepSeek V4 Pro (1.6T MoE) on Apple M1 Ultra: not started
Qwen 3 235B-A22B on Apple M1 Ultra: not started
Nomic Embed Text v1.5 on Apple M1 Ultra: not started
Kokoro 82M on Apple M1 Ultra: not started
Llama 4 Scout on Apple M1 Ultra: not started
Llama 3.1 8B Instruct on Apple M1 Ultra: not started
DeepSeek V4 Flash (284B MoE) on Apple M1 Ultra: not started
DeepSeek R1 (671B reasoning) on Apple M1 Ultra: not started
BGE Large EN v1.5 on Apple M1 Ultra: not started
Qwen 3 0.6B on Apple M1 Ultra: not started
Qwen 3 30B-A3B on Apple M1 Ultra: not started
Llama 3.3 70B Instruct on Apple M1 Ultra: not started
AMD Instinct MI300X
all-MiniLM-L6-v2 on AMD Instinct MI300X: not started
FLUX.1 [dev] on AMD Instinct MI300X: not started
DeepSeek V4 Pro (1.6T MoE) on AMD Instinct MI300X: not started
Qwen 3 235B-A22B on AMD Instinct MI300X: not started
Nomic Embed Text v1.5 on AMD Instinct MI300X: not started
Kokoro 82M on AMD Instinct MI300X: not started
Llama 4 Scout on AMD Instinct MI300X: not started
Llama 3.1 8B Instruct on AMD Instinct MI300X: not started
DeepSeek V4 Flash (284B MoE) on AMD Instinct MI300X: not started
DeepSeek R1 (671B reasoning) on AMD Instinct MI300X: not started
BGE Large EN v1.5 on AMD Instinct MI300X: not started
Qwen 3 0.6B on AMD Instinct MI300X: not started
Qwen 3 30B-A3B on AMD Instinct MI300X: not started
Llama 3.3 70B Instruct on AMD Instinct MI300X: not started
Apple M2 Ultra
all-MiniLM-L6-v2 on Apple M2 Ultra: not started
FLUX.1 [dev] on Apple M2 Ultra: not started
DeepSeek V4 Pro (1.6T MoE) on Apple M2 Ultra: not started
Qwen 3 235B-A22B on Apple M2 Ultra: not started
Nomic Embed Text v1.5 on Apple M2 Ultra: not started
Kokoro 82M on Apple M2 Ultra: not started
Llama 4 Scout on Apple M2 Ultra: not started
Llama 3.1 8B Instruct on Apple M2 Ultra: not started
DeepSeek V4 Flash (284B MoE) on Apple M2 Ultra: not started
DeepSeek R1 (671B reasoning) on Apple M2 Ultra: not started
BGE Large EN v1.5 on Apple M2 Ultra: not started
Qwen 3 0.6B on Apple M2 Ultra: not started
Qwen 3 30B-A3B on Apple M2 Ultra: not started
Llama 3.3 70B Instruct on Apple M2 Ultra: not started
Apple A17 Pro
all-MiniLM-L6-v2 on Apple A17 Pro: not started
FLUX.1 [dev] on Apple A17 Pro: not started
DeepSeek V4 Pro (1.6T MoE) on Apple A17 Pro: not started
Qwen 3 235B-A22B on Apple A17 Pro: not started
Nomic Embed Text v1.5 on Apple A17 Pro: not started
Kokoro 82M on Apple A17 Pro: not started
Llama 4 Scout on Apple A17 Pro: not started
Llama 3.1 8B Instruct on Apple A17 Pro: not started
DeepSeek V4 Flash (284B MoE) on Apple A17 Pro: not started
DeepSeek R1 (671B reasoning) on Apple A17 Pro: not started
BGE Large EN v1.5 on Apple A17 Pro: not started
Qwen 3 0.6B on Apple A17 Pro: not started
Qwen 3 30B-A3B on Apple A17 Pro: not started
Llama 3.3 70B Instruct on Apple A17 Pro: not started
Apple A18 Pro
all-MiniLM-L6-v2 on Apple A18 Pro: not started
FLUX.1 [dev] on Apple A18 Pro: not started
DeepSeek V4 Pro (1.6T MoE) on Apple A18 Pro: not started
Qwen 3 235B-A22B on Apple A18 Pro: not started
Nomic Embed Text v1.5 on Apple A18 Pro: not started
Kokoro 82M on Apple A18 Pro: not started
Llama 4 Scout on Apple A18 Pro: not started
Llama 3.1 8B Instruct on Apple A18 Pro: not started
DeepSeek V4 Flash (284B MoE) on Apple A18 Pro: not started
DeepSeek R1 (671B reasoning) on Apple A18 Pro: not started
BGE Large EN v1.5 on Apple A18 Pro: not started
Qwen 3 0.6B on Apple A18 Pro: not started
Qwen 3 30B-A3B on Apple A18 Pro: not started
Llama 3.3 70B Instruct on Apple A18 Pro: not started
Apple M1 Max
all-MiniLM-L6-v2 on Apple M1 Max: not started
FLUX.1 [dev] on Apple M1 Max: not started
DeepSeek V4 Pro (1.6T MoE) on Apple M1 Max: not started
Qwen 3 235B-A22B on Apple M1 Max: not started
Nomic Embed Text v1.5 on Apple M1 Max: not started
Kokoro 82M on Apple M1 Max: not started
Llama 4 Scout on Apple M1 Max: not started
Llama 3.1 8B Instruct on Apple M1 Max: not started
DeepSeek V4 Flash (284B MoE) on Apple M1 Max: not started
DeepSeek R1 (671B reasoning) on Apple M1 Max: not started
BGE Large EN v1.5 on Apple M1 Max: not started
Qwen 3 0.6B on Apple M1 Max: not started
Qwen 3 30B-A3B on Apple M1 Max: not started
Llama 3.3 70B Instruct on Apple M1 Max: not started
Apple M2 Max
all-MiniLM-L6-v2 on Apple M2 Max: not started
FLUX.1 [dev] on Apple M2 Max: not started
DeepSeek V4 Pro (1.6T MoE) on Apple M2 Max: not started
Qwen 3 235B-A22B on Apple M2 Max: not started
Nomic Embed Text v1.5 on Apple M2 Max: not started
Kokoro 82M on Apple M2 Max: not started
Llama 4 Scout on Apple M2 Max: not started
Llama 3.1 8B Instruct on Apple M2 Max: not started
DeepSeek V4 Flash (284B MoE) on Apple M2 Max: not started
DeepSeek R1 (671B reasoning) on Apple M2 Max: not started
BGE Large EN v1.5 on Apple M2 Max: not started
Qwen 3 0.6B on Apple M2 Max: not started
Qwen 3 30B-A3B on Apple M2 Max: not started
Llama 3.3 70B Instruct on Apple M2 Max: not started
Apple M3 Ultra
all-MiniLM-L6-v2 on Apple M3 Ultra: not started
FLUX.1 [dev] on Apple M3 Ultra: not started
DeepSeek V4 Pro (1.6T MoE) on Apple M3 Ultra: not started
Qwen 3 235B-A22B on Apple M3 Ultra: not started
Nomic Embed Text v1.5 on Apple M3 Ultra: not started
Kokoro 82M on Apple M3 Ultra: not started
Llama 4 Scout on Apple M3 Ultra: not started
Llama 3.1 8B Instruct on Apple M3 Ultra: not started
DeepSeek V4 Flash (284B MoE) on Apple M3 Ultra: not started
DeepSeek R1 (671B reasoning) on Apple M3 Ultra: not started
BGE Large EN v1.5 on Apple M3 Ultra: not started
Qwen 3 0.6B on Apple M3 Ultra: not started
Qwen 3 30B-A3B on Apple M3 Ultra: not started
Llama 3.3 70B Instruct on Apple M3 Ultra: not started
Apple M4 (iPad Pro)
all-MiniLM-L6-v2 on Apple M4 (iPad Pro): not started
FLUX.1 [dev] on Apple M4 (iPad Pro): not started
DeepSeek V4 Pro (1.6T MoE) on Apple M4 (iPad Pro): not started
Qwen 3 235B-A22B on Apple M4 (iPad Pro): not started
Nomic Embed Text v1.5 on Apple M4 (iPad Pro): not started
Kokoro 82M on Apple M4 (iPad Pro): not started
Llama 4 Scout on Apple M4 (iPad Pro): not started
Llama 3.1 8B Instruct on Apple M4 (iPad Pro): not started
DeepSeek V4 Flash (284B MoE) on Apple M4 (iPad Pro): not started
DeepSeek R1 (671B reasoning) on Apple M4 (iPad Pro): not started
BGE Large EN v1.5 on Apple M4 (iPad Pro): not started
Qwen 3 0.6B on Apple M4 (iPad Pro): not started
Qwen 3 30B-A3B on Apple M4 (iPad Pro): not started
Llama 3.3 70B Instruct on Apple M4 (iPad Pro): not started
Apple M4 Max
all-MiniLM-L6-v2 on Apple M4 Max: not started
FLUX.1 [dev] on Apple M4 Max: not started
DeepSeek V4 Pro (1.6T MoE) on Apple M4 Max: not started
Qwen 3 235B-A22B on Apple M4 Max: not started
Nomic Embed Text v1.5 on Apple M4 Max: not started
Kokoro 82M on Apple M4 Max: not started
Llama 4 Scout on Apple M4 Max: not started
Llama 3.1 8B Instruct on Apple M4 Max: not started
DeepSeek V4 Flash (284B MoE) on Apple M4 Max: not started
DeepSeek R1 (671B reasoning) on Apple M4 Max: not started
BGE Large EN v1.5 on Apple M4 Max: not started
Qwen 3 0.6B on Apple M4 Max: not started
Qwen 3 30B-A3B on Apple M4 Max: not started
Llama 3.3 70B Instruct on Apple M4 Max: not started
Apple M4 Pro
all-MiniLM-L6-v2 on Apple M4 Pro: not started
FLUX.1 [dev] on Apple M4 Pro: not started
DeepSeek V4 Pro (1.6T MoE) on Apple M4 Pro: not started
Qwen 3 235B-A22B on Apple M4 Pro: not started
Nomic Embed Text v1.5 on Apple M4 Pro: not started
Kokoro 82M on Apple M4 Pro: not started
Llama 4 Scout on Apple M4 Pro: not started
Llama 3.1 8B Instruct on Apple M4 Pro: not started
DeepSeek V4 Flash (284B MoE) on Apple M4 Pro: not started
DeepSeek R1 (671B reasoning) on Apple M4 Pro: not started
BGE Large EN v1.5 on Apple M4 Pro: not started
Qwen 3 0.6B on Apple M4 Pro: not started
Qwen 3 30B-A3B on Apple M4 Pro: not started
Llama 3.3 70B Instruct on Apple M4 Pro: not started
RunLocalAI measuredReproducedVendor-publishedCommunity-reviewedLow/unverified measuredEstimatedCritical wantedWantedNot started

No benchmarks published yet.

Owner-run measurements will appear here as they're added through the admin form.