New Benchmark Analysis Report: Deep dive into 1,700+ community runs across 15 Apple Silicon chips with interactive charts

Anubis Leaderboard

Real-world LLM performance on Apple Silicon: tok/s, power, TTFT. Run a benchmark in Anubis and upload your results.

Anubis
The Architect’s Toolkit BUNDLE
devPad + cyberWriter + Anubis
cyberWriter
cyberWriter 25% OFF
Anubis user discount, code applied
W_
Wranglify DATA
Local data workbench, nothing uploads
Total Runs
0
community submissions
Fastest tok/s
0
 
Most-Tested Chip
 
 
Unique Models
0
distinct LLMs benchmarked
Loading…
Wildcards  qwen3* starts with · *coder* contains · *-30b-* anywhere · llama?2 single character Combine  qwen 30b all terms must match · mlx|gguf either · -1b exclude Matches  model name, model id, quantization, format, chip, backend and submitter. Plain text matches anywhere; wildcards match a whole field.
Full dataset 0 runs × 59 fields

Loading leaderboard…

Note on coverage: only runs submitted from an Ollama backend carry the full gamut of datapoints. LM Studio, MLX and others do not expose every performance metric, such as prompt evaluation time. The dataset is robust even without Ollama as the runtime. Enjoy! –JT

More from uncSoft

All native, all local-first, none of them phone home.

Small, sandboxed macOS tools built by one person. Anubis is the free and open-source one; the others fund it.