What each lane serves on each machine, the receipt behind it, and the latest run's full comparison.
| machine | last published | age | schema | export | |
|---|---|---|---|---|---|
| Mac Studio M5 Ultra 96 GB | 2026-11-12T04:30:40Z | 0.0d | 98 | 4 | current |
Each machine publishes its own measurements. One that has not published in 7 days is marked stale, and its numbers are as old as its publish date.
| lane | Mac Studio M5 Ultra 96 GB |
|---|---|
| image | mflux:dev 0.99 pass, 28.91s, 2026-11-12 |
| code | llamacpp:Ornith-1.5-35B-Q4_K_M; 35B (3B active) params, Q4_K_M, by ornith-ai 0.79 pass, 3.04s, 2026-11-12 |
| web | mlx-community/Qwen3-4B-Instruct-2507-4bit (served as q3-4b); 4B params, 4bit, by mlx-community 0.64 pass, 2.94s, 2026-11-12 |
| svg | mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit (served as q3-30b); 30B (3B active) params, 4bit, by mlx-community 0.72 pass, 4.53s, 2026-11-11 |
| video | h3 1.00 pass, 171.53s, 2026-11-09 |
| music | acestep:acestep-v15-turbo 1.00 pass, 28.42s, 2026-11-02 stale |
| decide | imajev-4b-Q8_0; 4B params, Q8_0, by mindchain not run (harness_error) |
| agent | llamacpp:Ornith-1.5-35B-Q4_K_M; 35B (3B active) params, Q4_K_M, by ornith-ai 0.95 pass, 9.17s, 2026-11-05 stale |
| ocr | hf-ocr:PaddlePaddle/PaddleOCR-VL-1.6 0.83 pass, 7.64s, 2026-11-11 |
| retrieval | bm25 0.78 pass, 0.02s, 2026-11-10 |
| pii | pii-regex 0.74 pass, 0.03s, 2026-11-09 |
| claims | Qwen2.5-7B-Instruct-Q4_K_M; 7B params, Q4_K_M, by bartowski claims_balanced 0.593, pass 192/299 0.64, 0.14s, 2026-11-02 stale |
| repo | llamacpp:Ornith-1.5-35B-Q4_K_M; 35B (3B active) params, Q4_K_M, by ornith-ai resolved 0.100, pass 1/10 0.10, 201.81s, 2026-11-07 |
| extract | mlx-community/Qwen2.5-7B-Instruct-4bit; 7B params, 4bit, by mlx-community 0.72 pass, 0.13s, 2026-10-22 stale |
| stt | mlx-community/parakeet-tdt-0.6b-v2; 0.6B params, by mlx-community wer 0.015, pass 473/480 0.98, 0.09s, 2026-11-09 |
| tts | mlx-community/Kokoro-82M-bf16; family yl4579/StyleTTS2-LJSpeech, 82M params, bf16, by mlx-community 0.85 pass, 0.16s, 2026-11-09 |
Wall-clock is only comparable within one machine. Across machines compare pass rates and what each lane serves.
macOS 27.2 · arm64 · cpu, ds4, llamacpp, mlx · published 2026-11-12T04:30:40Z · schema 98
| lane | serves | pass | median s | TTFT med / p95 | metric | measured | |
|---|---|---|---|---|---|---|---|
| image | mflux:dev | 0.99 | 28.91 | -- | staging 0.988 | 2026-11-12 | adopted by-hand |
| code | llamacpp:Ornith-1.5-35B-Q4_K_M; 35B (3B active) params, Q4_K_M, by ornith-ai | 0.79 | 3.04 | 2.58s / 18.35s | co_tenant 1.000 code_pass 0.833 completion_tokens 983.375 tokens_per_s 128.725 | 2026-11-12 | adopted measured |
| web | mlx-community/Qwen3-4B-Instruct-2507-4bit (served as q3-4b); 4B params, 4bit, by mlx-community | 0.64 | 2.94 | -- | completion_tokens 610.774 ink 0.084 method_calls 1.810 method_chosen 1.041 method_samples 3.000 | 2026-11-12 | adopted measured |
| svg | mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit (served as q3-30b); 30B (3B active) params, 4bit, by mlx-community | 0.72 | 4.53 | -- | completion_tokens 567.452 ink 0.385 method_calls 2.000 svg_bytes 654.382 | 2026-11-11 | adopted measured |
| video | h3 | 1.00 | 171.53 | -- | motion 0.038 | 2026-11-09 | |
| music | acestep:acestep-v15-turbo | 1.00 | 28.42 | -- | wer 0.339 duration_error_s 0.000 seconds 30.000 | 2026-11-02 | stale |
| decide | imajev-4b-Q8_0; 4B params, Q8_0, by mindchain | decide_accuracy --, all fields right -- | -- | -- | 2026-11-10 | not run (harness_error) | |
| agent | llamacpp:Ornith-1.5-35B-Q4_K_M; 35B (3B active) params, Q4_K_M, by ornith-ai | 0.95 | 9.17 | 0.30s / 0.42s | agent_ctx 262144.000 agent_model_s 11.044 agent_no_tool_steps 0.000 agent_steps 8.265 agent_tool_s 0.151 agent_ttft_sum_s 5.444 agent_valid_call_rate 1.000 agent_wall_s 11.259 completion_tokens 1435.979 prompt_tokens 14245.134 | 2026-11-05 | adopted measured stale |
| ocr | hf-ocr:PaddlePaddle/PaddleOCR-VL-1.6 | 0.83 | 7.64 | -- | cer 0.646 | 2026-11-11 | adopted measured |
| retrieval | bm25 | 0.78 | 0.02 | -- | retrieval_ndcg 0.880 retrieval_recall 0.783 | 2026-11-10 | |
| pii | pii-regex | 0.74 | 0.03 | -- | pii_f1 0.796 pii_recall 0.662 pii_precision 1.000 | 2026-11-09 | |
| claims | Qwen2.5-7B-Instruct-Q4_K_M; 7B params, Q4_K_M, by bartowski | claims_balanced 0.593, pass 192/299 0.64 | 0.14 | 0.08s / 0.20s | claims_balanced 0.593 claims_case_recall 0.521 claims_case_specificity 0.665 claims_empty_rate 0.668 claims_schema_valid 1.000 claims_bad_matched_cited_only 0.091 claims_bad_matched_conversation 0.000 claims_emitted 0.875 claims_judge_calibrated 0.000 claims_matcher_version 2.000 claims_near_miss_cited_only 0.182 claims_near_miss_conversation 0.067 claims_pass_cited_only 0.318 claims_pass_conversation 0.800 claims_precision_cited_only 0.821 claims_precision_conversation 0.922 claims_recall_cited_only 0.287 claims_recall_conversation 0.657 claims_unreviewed_cited_only 1.727 claims_unreviewed_conversation 1.467 claims_verbatim_cited_only 0.273 claims_verbatim_conversation 0.978 completion_tokens 27.260 tokens_per_s 65.195 | 2026-11-02 | stale |
| repo | llamacpp:Ornith-1.5-35B-Q4_K_M; 35B (3B active) params, Q4_K_M, by ornith-ai | resolved 0.100, pass 1/10 0.10 | 201.81 | 116.81s / 116.81s | resolved 0.100 patch_applied 0.100 completion_tokens 11054.000 test_s 122.386 tokens_per_s 86.700 wall_s 250.780 | 2026-11-07 | adopted measured |
| extract | mlx-community/Qwen2.5-7B-Instruct-4bit; 7B params, 4bit, by mlx-community | 0.72 | 0.13 | 0.10s / 0.14s | completion_tokens 6.649 tokens_per_s 40.871 | 2026-10-22 | stale |
| stt | mlx-community/parakeet-tdt-0.6b-v2; 0.6B params, by mlx-community | wer 0.015, pass 473/480 0.98 | 0.09 | -- | wer 0.015 | 2026-11-09 | |
| tts | mlx-community/Kokoro-82M-bf16; family yl4579/StyleTTS2-LJSpeech, 82M params, bf16, by mlx-community | 0.85 | 0.16 | -- | wer 0.097 | 2026-11-09 |
image latest run 2026-11-12
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| mflux/dev-q8 | 80/81 | 0.99 | 28.91 | -- | 20.6 | staging 0.988 |
code latest run 2026-11-12
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| llamacpp:Ornith-1.5-35B-Q4_K_M 35B (3B active) params, Q4_K_M, by ornith-ai source | 19/24 | 0.79 | 3.04 | 2.58s / 18.35s | 0.0 | co_tenant 1.000 code_pass 0.833 completion_tokens 983.375 tokens_per_s 128.725 |
| llamacpp:Qwen3.6-35B-A3B-UD-Q4_K_M family Qwen/Qwen3.6-35B-A3B, 35B (3B active) params, Q4_K_M, by unsloth source | 18/23 | 0.78 | 32.28 | 28.08s / 54.67s | 0.0 | code_pass 0.864 completion_tokens 3796.727 tokens_per_s 130.673 |
web latest run 2026-11-12
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| best-of-3/q3-4b: mlx-community/Qwen3-4B-Instruct-2507-4bit 4B params, 4bit, by mlx-community source | 124/195 | 0.64 | 2.94 | -- | 0.0 | completion_tokens 610.774 ink 0.084 method_calls 1.810 method_chosen 1.041 method_samples 3.000 |
| mlx-community/Qwen3-4B-Instruct-2507-4bit 4B params, 4bit, by mlx-community source | 119/194 | 0.61 | 2.21 | 0.13s / 0.16s | 0.0 | completion_tokens 330.299 ink 0.075 tokens_per_s 117.028 |
svg latest run 2026-11-11
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| plan/q3-30b: mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit 30B (3B active) params, 4bit, by mlx-community source | 454/628 | 0.72 | 4.53 | -- | 0.0 | completion_tokens 567.452 ink 0.385 method_calls 2.000 svg_bytes 654.382 |
| mlx-community/Qwen3-30B-A3B-Instruct-2507-4bit (served as q3-30b) 30B (3B active) params, 4bit, by mlx-community source | 434/628 | 0.69 | 2.45 | 0.15s / 0.18s | 0.0 | completion_tokens 328.292 ink 0.326 svg_bytes 591.624 tokens_per_s 113.984 |
video latest run 2026-11-09
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| diffusers-video/Motif-Video-2B 2B params | 2/2 | 1.00 | 50.76 | -- | 21.5 | motion 0.083 |
| h3/minimax-h3 | 2/2 | 1.00 | 171.53 | -- | 3.5 | motion 0.038 |
music latest run 2026-11-02
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| acestep/acestep-v15-turbo | 4/4 | 1.00 | 28.42 | -- | 46.0 | wer 0.339 duration_error_s 0.000 seconds 30.000 |
decide latest run 2026-11-12
| candidate | decide_accuracy | all fields right | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|---|
| llamacpp:imajev-4b-Q8_0 4B params, Q8_0, by mindchain source | 0.823 | 241/529 | 0.46 | 0.46 | -- | 0.0 | decide_accuracy 0.823 decide_brier 0.258 decide_ece 0.030 calibrated 1.000 co_tenant 1.000 completion_tokens 36.461 tokens_per_s 77.903 |
| bespokelabs/Bespoke-Nimble-9B family Qwen/Qwen3.5-9B, 9B params, by bespokelabs source | not run (harness_error) | -- | -- | -- | |||
agent latest run 2026-11-05
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| llamacpp:Ornith-1.5-35B-Q4_K_M 35B (3B active) params, Q4_K_M, by ornith-ai source | 356/374 | 0.95 | 9.17 | 0.30s / 0.42s | 0.0 | agent_ctx 262144.000 agent_model_s 11.044 agent_no_tool_steps 0.000 agent_steps 8.265 agent_tool_s 0.151 agent_ttft_sum_s 5.444 agent_valid_call_rate 1.000 agent_wall_s 11.259 completion_tokens 1435.979 prompt_tokens 14245.134 |
| mlx-community/Qwen3-4B-Instruct-2507-4bit (served as q3-4b) 4B params, 4bit, by mlx-community source | 52/374 | 0.14 | 2.22 | 0.18s / 0.25s | 0.0 | agent_model_s 4.514 agent_no_tool_steps 0.024 agent_steps 5.660 agent_tool_s 0.028 agent_ttft_sum_s 1.500 agent_valid_call_rate 1.000 agent_wall_s 4.583 completion_tokens 667.869 prompt_tokens 6416.305 |
ocr latest run 2026-11-11
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| hf-ocr/PaddleOCR-VL-1.6 | 371/448 | 0.83 | 7.64 | -- | 0.0 | cer 0.646 |
| osocr/auto | 323/448 | 0.72 | 0.30 | -- | 0.2 | cer 0.170 |
retrieval latest run 2026-11-10
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| bm25/okapi | 148/189 | 0.78 | 0.02 | -- | 0.0 | retrieval_ndcg 0.880 retrieval_recall 0.783 |
pii latest run 2026-11-09
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| hf-pii/privacy-filter | 462/556 | 0.83 | 3.92 | -- | 4.0 | pii_f1 0.890 pii_recall 0.871 pii_precision 0.910 |
| pii-regex/patterns | 411/556 | 0.74 | 0.03 | -- | 0.0 | pii_f1 0.796 pii_recall 0.662 pii_precision 1.000 |
claims latest run 2026-11-02
| candidate | claims_balanced | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|---|
| Qwen2.5-7B-Instruct-Q4_K_M 7B params, Q4_K_M, by bartowski source | 0.593 | 192/299 | 0.64 | 0.14 | 0.08s / 0.20s | 0.0 | claims_balanced 0.593 claims_case_recall 0.521 claims_case_specificity 0.665 claims_empty_rate 0.668 claims_schema_valid 1.000 claims_bad_matched_cited_only 0.091 claims_bad_matched_conversation 0.000 claims_emitted 0.875 claims_judge_calibrated 0.000 claims_matcher_version 2.000 claims_near_miss_cited_only 0.182 claims_near_miss_conversation 0.067 claims_pass_cited_only 0.318 claims_pass_conversation 0.800 claims_precision_cited_only 0.821 claims_precision_conversation 0.922 claims_recall_cited_only 0.287 claims_recall_conversation 0.657 claims_unreviewed_cited_only 1.727 claims_unreviewed_conversation 1.467 claims_verbatim_cited_only 0.273 claims_verbatim_conversation 0.978 completion_tokens 27.260 tokens_per_s 65.195 |
repo latest run 2026-11-07
| candidate | resolved | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|---|
| llamacpp:Qwen3.6-27B-Q4_K_M family Qwen/Qwen3.6-27B, 27B params, Q4_K_M, by unsloth source | 0.200 | 1/5 | 0.20 | 592.50 | 280.09s / 524.58s | 0.0 | resolved 0.200 patch_applied 0.400 completion_tokens 13413.000 test_s 36.203 tokens_per_s 33.967 wall_s 424.803 |
| llamacpp:Ornith-1.5-35B-Q4_K_M 35B (3B active) params, Q4_K_M, by ornith-ai source | 0.100 | 1/10 | 0.10 | 201.81 | 116.81s / 116.81s | 0.0 | resolved 0.100 patch_applied 0.100 completion_tokens 11054.000 test_s 122.386 tokens_per_s 86.700 wall_s 250.780 |
extract latest run 2026-10-22
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| mlx-community/Qwen2.5-7B-Instruct-4bit 7B params, 4bit, by mlx-community source | 325/450 | 0.72 | 0.13 | 0.10s / 0.14s | 0.0 | completion_tokens 6.649 tokens_per_s 40.871 |
| plan/local-large: mlx-community/Qwen2.5-7B-Instruct-4bit 7B params, 4bit, by mlx-community source | 282/450 | 0.63 | 0.26 | -- | 0.0 | completion_tokens 19.693 method_calls 2.000 |
stt latest run 2026-11-09
| candidate | wer | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|---|
| Confucius4-R2T2 | 0.014 | 475/480 | 0.99 | 0.23 | -- | 0.0 | wer 0.014 |
| parakeet-tdt-0.6b-v2 0.6B params | 0.015 | 473/480 | 0.98 | 0.09 | -- | 0.0 | wer 0.015 |
tts latest run 2026-11-09
| candidate | pass | pass | median s | TTFT med / p95 | peak GB | metric |
|---|---|---|---|---|---|---|
| fish-audio-s2-pro | 398/465 | 0.86 | 3.33 | -- | 0.0 | wer 0.100 |
| Kokoro-82M-bf16/bm_george 82M params, bf16, by Kokoro-82M-bf16 source | 397/465 | 0.85 | 0.16 | -- | 0.0 | wer 0.097 |
Real use last 7 days through the gateway
| alias | requests | tokens in / out | TTFT p50 / p95 | errors | bad tool calls |
|---|---|---|---|---|---|
| sohot-agent | 17 | 289 / 34 | -- | 0.0% | -- |
| sohot-claims | 573 | 25124 / 55387 | -- | 0.3% | -- |
| sohot-code | 487 | 14832 / 8931 | 0.35s / 0.52s | 0.8% | -- |
| sohot-decide | 951 | 119357 / 10944 | 0.04s / 0.04s | 0.0% | -- |
| sohot-extract | 1191 | 494683 / 48614 | 0.59s / 0.70s | 0.0% | -- |
Adoptions
| lane | adopted | replaced | how | when |
|---|---|---|---|---|
| code | llamacpp:Ornith-1.5-35B-Q4_K_M | -- | measured | 2026-10-06 |
| image | mflux:dev | mflux:flux2-klein-4b | by-hand | 2026-10-06 |
| svg | q3-30b | local-large | measured | 2026-10-06 |
| agent | llamacpp:Ornith-1.5-35B-Q4_K_M | q3-4b | measured | 2026-10-07 |
| repo | llamacpp:Ornith-1.5-35B-Q4_K_M | mlx-community/Qwen3-4B-Instruct-2507-4bit | measured | 2026-10-24 |
| ocr | hf-ocr:PaddlePaddle/PaddleOCR-VL-1.6 | osocr:auto | measured | 2026-11-07 |
| svg | plan:q3-30b | q3-30b | measured | 2026-11-10 |
| web | best-of:3:q3-4b | mlx-community/Qwen3-4B-Instruct-2507-4bit | measured | 2026-11-11 |