Chips
Loading community benchmarks…
Loading community benchmarks…
Submitted benchmarks for Apple M3 Ultra. Browse individual runs and their measurement settings, or open the report JSON for full benchmark details.
Peak throughput is the highest reported value and may come from different runs. Compare model, quantisation, backend, and token counts before drawing conclusions. Model verification identifies published artifact bytes; community benchmark execution is not remotely attested.
Apple M3 Ultra Metal |
111.4 tok/s TG128 @ 1 ctx |
2,310 tok/s PP512 |
| 3,793.1 MiB |
basecompute |
| View Benchmark |
Qwen3.5-35B-A3B-Q4 BaseRTQ4 Nothing in this report ties it to a published model file. Reports from CLI 0.1.0 record only a model name and a file hash, so the run is grouped by the name it reported until it is reconciled. | Apple M3 Ultra Metal | 136.6 tok/s TG128 @ 1 ctx | 2,329 tok/s PP512 | 2,861.1 MiB | basecompute | View Benchmark |
Qwen3.6-27B-Q4 BaseRTQ4 Nothing in this report ties it to a published model file. Reports from CLI 0.1.0 record only a model name and a file hash, so the run is grouped by the name it reported until it is reconciled. | Apple M3 Ultra Metal | 38.3 tok/s TG128 @ 1 ctx | 410 tok/s PP512 | 22,670.3 MiB | basecompute | View Benchmark |
Qwen3.5-2B-Base-Q8 BaseRTQ8 Nothing in this report ties it to a published model file. Reports from CLI 0.1.0 record only a model name and a file hash, so the run is grouped by the name it reported until it is reconciled. | Apple M3 Ultra Metal | 228.6 tok/s TG128 @ 1 ctx | 5,433 tok/s PP512 | 2,953.8 MiB | basecompute | View Benchmark |
Qwen3.5-2B-Q8 BaseRTQ8 Nothing in this report ties it to a published model file. Reports from CLI 0.1.0 record only a model name and a file hash, so the run is grouped by the name it reported until it is reconciled. | Apple M3 Ultra Metal | 227.4 tok/s TG128 @ 1 ctx | 5,418 tok/s PP512 | 2,953.5 MiB | basecompute | View Benchmark |
Qwen3.5-2B-Q4 BaseRTQ4 Nothing in this report ties it to a published model file. Reports from CLI 0.1.0 record only a model name and a file hash, so the run is grouped by the name it reported until it is reconciled. | Apple M3 Ultra Metal | 313.4 tok/s TG128 @ 1 ctx | 5,478 tok/s PP512 | 2,068.7 MiB | basecompute | View Benchmark |
Qwen3.5-2B-Base-Q4 BaseRTQ4 Nothing in this report ties it to a published model file. Reports from CLI 0.1.0 record only a model name and a file hash, so the run is grouped by the name it reported until it is reconciled. | Apple M3 Ultra Metal | 312.3 tok/s TG128 @ 1 ctx | 5,495 tok/s PP512 | 2,068.9 MiB | basecompute | View Benchmark |
gemma-3-1b-it-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 331.4 tok/s TG128 @ 1 ctx | 10,125 tok/s PP512 | 941.3 MiB | basecompute | View Benchmark |
Qwen3-0.6B-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 488.1 tok/s TG128 @ 1 ctx | 12,301 tok/s PP512 | 1,375.5 MiB | basecompute | View Benchmark |
gemma-3-1b-it-Q8 BaseRTQ8 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 262.3 tok/s TG128 @ 1 ctx | 9,865 tok/s PP512 | 1,367.9 MiB | basecompute | View Benchmark |
Qwen3-0.6B-Q8 BaseRTQ8 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 418.6 tok/s TG128 @ 1 ctx | 11,505 tok/s PP512 | 1,648.5 MiB | basecompute | View Benchmark |
Llama-3.2-1B-Instruct-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 415.1 tok/s TG128 @ 1 ctx | 8,900 tok/s PP512 | 1,090.8 MiB | basecompute | View Benchmark |
Qwen3-1.7B-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 314.9 tok/s TG128 @ 1 ctx | 5,751 tok/s PP512 | 2,086.5 MiB | basecompute | View Benchmark |
Qwen3.5-2B-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 311.3 tok/s TG128 @ 1 ctx | 5,484 tok/s PP512 | 2,069.0 MiB | basecompute | View Benchmark |
Llama-3.2-1B-Instruct-Q8 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 356.8 tok/s TG128 @ 1 ctx | 9,006 tok/s PP512 | 1,675.7 MiB | basecompute | View Benchmark |
Qwen3.5-2B-Base-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 305.2 tok/s TG128 @ 1 ctx | 5,442 tok/s PP512 | 2,068.9 MiB | basecompute | View Benchmark |
Qwen3-1.7B-Q8 BaseRTQ8 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 246.8 tok/s TG128 @ 1 ctx | 5,666 tok/s PP512 | 2,941.3 MiB | basecompute | View Benchmark |
Qwen3-4B-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 170.9 tok/s TG128 @ 1 ctx | 2,591 tok/s PP512 | 3,899.5 MiB | basecompute | View Benchmark |
Qwen3.5-2B-Q8 BaseRTQ8 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 232.5 tok/s TG128 @ 1 ctx | 5,415 tok/s PP512 | 2,954.0 MiB | basecompute | View Benchmark |
Qwen3.5-2B-Base-Q8 BaseRTQ8 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 230.2 tok/s TG128 @ 1 ctx | 5,282 tok/s PP512 | 2,954.1 MiB | basecompute | View Benchmark |
Llama-3.2-3B-Instruct-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 203.1 tok/s TG128 @ 1 ctx | 3,306 tok/s PP512 | 3,090.1 MiB | basecompute | View Benchmark |
Qwen3-4B-Instruct-2507-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 160.2 tok/s TG128 @ 1 ctx | 2,541 tok/s PP512 | 3,899.7 MiB | basecompute | View Benchmark |
Qwen3-4B-Thinking-2507-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 161.0 tok/s TG128 @ 1 ctx | 2,546 tok/s PP512 | 3,899.8 MiB | basecompute | View Benchmark |
gemma-4-E2B-it-Q4 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 142.7 tok/s TG128 @ 1 ctx | 8,609 tok/s PP512 | 1,587.0 MiB | basecompute | View Benchmark |
Llama-3.2-3B-Instruct-Q8 BaseRTQ4 The report's model SHA-256 matches a file published on Hugging Face, so the exact model bytes are known. This does not attest benchmark execution. | Apple M3 Ultra Metal | 159.9 tok/s TG128 @ 1 ctx | 3,309 tok/s PP512 | 4,773.1 MiB | basecompute | View Benchmark |