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 |
122.5 tok/s TG128 @ 1 ctx |
2,542 tok/s PP512 |
| 6,038.5 MiB |
basecompute |
| View Benchmark |
Qwen3-4B-Instruct-2507-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 | 122.0 tok/s TG128 @ 1 ctx | 2,550 tok/s PP512 | 6,038.5 MiB | basecompute | View Benchmark |
Qwen3-4B-Thinking-2507-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 | 121.5 tok/s TG128 @ 1 ctx | 2,548 tok/s PP512 | 6,038.3 MiB | basecompute | View Benchmark |
Mistral-7B-Instruct-v0.3-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 | 131.3 tok/s TG128 @ 1 ctx | 1,435 tok/s PP512 | 6,117.7 MiB | basecompute | View Benchmark |
gemma-4-E4B-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 | 98.8 tok/s TG128 @ 1 ctx | 3,540 tok/s PP512 | 3,328.3 MiB | basecompute | View Benchmark |
Qwen3-8B-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 | 118.3 tok/s TG128 @ 1 ctx | 1,398 tok/s PP512 | 6,260.6 MiB | basecompute | View Benchmark |
Llama-3.1-8B-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 | 125.7 tok/s TG128 @ 1 ctx | 1,430 tok/s PP512 | 6,178.8 MiB | basecompute | View Benchmark |
gemma-4-E2B-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 | 122.4 tok/s TG128 @ 1 ctx | 8,469 tok/s PP512 | 2,613.2 MiB | basecompute | View Benchmark |
Mistral-7B-Instruct-v0.3-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 | 81.7 tok/s TG128 @ 1 ctx | 1,427 tok/s PP512 | 10,261.3 MiB | basecompute | View Benchmark |
Llama-3.1-8B-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 | 78.1 tok/s TG128 @ 1 ctx | 1,423 tok/s PP512 | 10,322.4 MiB | basecompute | View Benchmark |
gemma-4-E4B-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 | 78.0 tok/s TG128 @ 1 ctx | 3,503 tok/s PP512 | 5,527.7 MiB | basecompute | View Benchmark |
Qwen3-8B-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 | 75.9 tok/s TG128 @ 1 ctx | 1,375 tok/s PP512 | 10,366.8 MiB | basecompute | View Benchmark |
Qwen3.6-27B-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 | 38.4 tok/s TG128 @ 1 ctx | 409 tok/s PP512 | 22,671.2 MiB | basecompute | View Benchmark |
gemma-4-26B-A4B-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 | 84.5 tok/s TG128 @ 1 ctx | 2,404 tok/s PP512 | 3,452.5 MiB | basecompute | View Benchmark |
Qwen3-30B-A3B-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 | 137.8 tok/s TG128 @ 1 ctx | 2,330 tok/s PP512 | 1,464.1 MiB | basecompute | View Benchmark |
Qwen3-30B-A3B-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 | 139.2 tok/s TG128 @ 1 ctx | 2,315 tok/s PP512 | 1,471.7 MiB | basecompute | View Benchmark |
Qwen3.5-35B-A3B-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 | 136.1 tok/s TG128 @ 1 ctx | 2,311 tok/s PP512 | 2,752.0 MiB | basecompute | View Benchmark |
gemma-4-26B-A4B-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 | 76.7 tok/s TG128 @ 1 ctx | 2,306 tok/s PP512 | 3,956.8 MiB | basecompute | View Benchmark |
Qwen3-30B-A3B-Thinking-2507-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 | 99.3 tok/s TG128 @ 1 ctx | 2,289 tok/s PP512 | 1,918.7 MiB | basecompute | View Benchmark |
Qwen3.5-35B-A3B-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 | 113.4 tok/s TG128 @ 1 ctx | 2,312 tok/s PP512 | 3,683.7 MiB | basecompute | View Benchmark |
gpt-oss-120b-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 | 121.4 tok/s TG128 @ 1 ctx | 1,695 tok/s PP512 | 9,208.2 MiB | basecompute | View Benchmark |
gpt-oss-120b-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 | 129.4 tok/s TG128 @ 1 ctx | 1,646 tok/s PP512 | 8,686.8 MiB | basecompute | View Benchmark |
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 BaseRTbf16 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 | 71.8 tok/s TG128 @ 1 ctx | 691 tok/s PP512 | 123,526.8 MiB | basecompute | View Benchmark |
gpt-oss-20b-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 | 185.1 tok/s TG128 @ 1 ctx | 2,554 tok/s PP512 | 7,877.6 MiB | basecompute | View Benchmark |
gpt-oss-20b-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 | 175.8 tok/s TG128 @ 1 ctx | 2,627 tok/s PP512 | 8,228.0 MiB | basecompute | View Benchmark |