[{"condition_id":"matrix-00","condition":"Qwen3.8-Flash-Next-REAP-288 · Q4g64 (30cmtonyfaker) · MLX-Serve · AR","baseline":"matrix-16","paired_tasks":["invoice","schedule","sql","extract","policy"],"n":5,"time_percent_arithmetic_mean":169.07523950293285,"time_percent_geometric_mean":140.39080753540162,"paired_speedup":0.7122973487760836,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-01","condition":"Qwen3.8-Flash-Next-REAP-288 · Q4g64 (30cmtonyfaker) · MLX-Serve · MTP","baseline":"matrix-16","paired_tasks":["invoice","schedule","extract","policy"],"n":4,"time_percent_arithmetic_mean":93.8782112517677,"time_percent_geometric_mean":85.41021168368137,"paired_speedup":1.1708201868220658,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-02","condition":"Qwen3.8-Flash-Next-REAP-288 · Q4g32 + Q8 (custom) · MTPLX native · AR","baseline":"matrix-16","paired_tasks":["invoice","schedule","intervals","sql","cache","extract","policy"],"n":7,"time_percent_arithmetic_mean":131.63042481636026,"time_percent_geometric_mean":112.68305382288248,"paired_speedup":0.8874448872958487,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-03","condition":"Qwen3.8-Flash-Next-REAP-288 · Q4g32 + Q8 (custom) · MTPLX native · MTP","baseline":"matrix-16","paired_tasks":["invoice","schedule","sql","cache","extract","policy"],"n":6,"time_percent_arithmetic_mean":85.05817338346054,"time_percent_geometric_mean":79.64242380424919,"paired_speedup":1.2556122129806988,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-04","condition":"Qwen3.8-Flash-Next-REAP-288 · Q4g64 (30cmtonyfaker) · MTPLX HTTP · AR","baseline":"matrix-16","paired_tasks":["policy","invoice","cache","extract"],"n":4,"time_percent_arithmetic_mean":141.1864880566946,"time_percent_geometric_mean":134.8812394147106,"paired_speedup":0.7413929500791173,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-05","condition":"Qwen3.8-Flash-Next-REAP-288 · Q4g64 (30cmtonyfaker) · MTPLX HTTP · MTP","baseline":"matrix-16","paired_tasks":["policy","invoice","schedule","cache","extract"],"n":5,"time_percent_arithmetic_mean":105.82046833408396,"time_percent_geometric_mean":98.37800505804965,"paired_speedup":1.016487373788412,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-08","condition":"Qwen 3.8 27B Q8 · MTPLX CLI historical · MTP","baseline":"matrix-16","paired_tasks":["schedule","intervals","batches","sql","cache","extract","policy","invoice"],"n":8,"time_percent_arithmetic_mean":78.74941442974931,"time_percent_geometric_mean":78.42835429362933,"paired_speedup":1.2750490674024373,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-09","condition":"Flash-Next 3.3 bpw · MLX-Serve · AR","baseline":"matrix-16","paired_tasks":["invoice","schedule","cache","extract","policy"],"n":5,"time_percent_arithmetic_mean":85.81144632443689,"time_percent_geometric_mean":81.9273296649112,"paired_speedup":1.2205939142531235,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-10","condition":"Flash-Next 3.3 bpw · MLX-Serve · MTP requested","baseline":"matrix-16","paired_tasks":["invoice","schedule","sql","cache","policy"],"n":5,"time_percent_arithmetic_mean":76.50389376017858,"time_percent_geometric_mean":70.63202271433644,"paired_speedup":1.4157884222633572,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-11","condition":"Gemma 26B-A4B Q8 · native MLX · AR","baseline":"matrix-16","paired_tasks":["invoice","schedule","intervals","batches","sql","extract","policy"],"n":7,"time_percent_arithmetic_mean":119.4557875269096,"time_percent_geometric_mean":117.2350049782693,"paired_speedup":0.8529875528092996,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-12","condition":"Gemma 26B-A4B Q8 · MLX-Serve · assistant off","baseline":"matrix-16","paired_tasks":["invoice","schedule","intervals","sql","extract","policy"],"n":6,"time_percent_arithmetic_mean":108.81763630019823,"time_percent_geometric_mean":104.13069768194731,"paired_speedup":0.9603316046670123,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-13","condition":"Gemma 26B-A4B Q8 · MLX-Serve · assistant on","baseline":"matrix-16","paired_tasks":["invoice","schedule","intervals","batches","extract","policy"],"n":6,"time_percent_arithmetic_mean":206.39018407292053,"time_percent_geometric_mean":194.67564549554308,"paired_speedup":0.5136749373320527,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-14","condition":"Qwen 3.8 27B Q4 · Youssofal MTPLX CLI historical · MTP","baseline":"matrix-16","paired_tasks":["invoice","schedule","intervals","batches","sql","cache","extract","policy"],"n":8,"time_percent_arithmetic_mean":79.4747731480297,"time_percent_geometric_mean":77.71820283511889,"paired_speedup":1.286699850897897,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-15","condition":"Swift Qwen 3.8 27B Q8 · MTPLX HTTP · medium · MTP","baseline":"matrix-16","paired_tasks":["invoice","schedule","intervals","batches","sql","cache","extract","policy"],"n":8,"time_percent_arithmetic_mean":98.33569455664474,"time_percent_geometric_mean":91.40995304114374,"paired_speedup":1.093972775098023,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-16","condition":"Qwen 3.8 27B Q8 · MTPLX HTTP · medium · MTP","baseline":"matrix-16","paired_tasks":["invoice","schedule","intervals","batches","sql","cache","extract","policy"],"n":8,"time_percent_arithmetic_mean":100.0,"time_percent_geometric_mean":100.0,"paired_speedup":1.0,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-17","condition":"Qwen 3.8 Flash · OpenRouter Alibaba · Hosted · reasoning budget32k","baseline":"matrix-16","paired_tasks":["invoice","schedule","intervals","batches","sql","extract","policy"],"n":7,"time_percent_arithmetic_mean":88.19005473778365,"time_percent_geometric_mean":54.08754884535954,"paired_speedup":1.848854350673344,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-18","condition":"Qwen 3.8 Flash · OpenRouter Alibaba · Hosted · reasoning budget64k","baseline":"matrix-16","paired_tasks":["invoice","schedule","intervals","batches","sql","cache","extract","policy"],"n":8,"time_percent_arithmetic_mean":44.910876139195786,"time_percent_geometric_mean":37.73774292597666,"paired_speedup":2.6498670096977452,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."},{"condition_id":"matrix-19","condition":"Qwen 3.8 27B BF16 · OpenRouter DeepInfra · Hosted · reasoning budget64k","baseline":"matrix-16","paired_tasks":["invoice","intervals","batches","cache","extract","policy"],"n":6,"time_percent_arithmetic_mean":35.896617056884175,"time_percent_geometric_mean":35.44044848861538,"paired_speedup":2.8216347214715194,"warning":"Different task subsets: do not rank these across models without selecting a common task intersection. Workflows are not fully matched."}]