{"schemaVersion":"2026-09-14","name":"Mac Terminal-Bench and Hugging Face artifact index with external Artificial Analysis references","capturedAt":"2026-09-14T14:43:34Z","records":[{"model":"GLM-5.3","organization":"Z.ai","exactOpenWeightMatch":true,"hf":{"baseRepo":"https://huggingface.co/zai-org/GLM-5.3","ggufRepo":"https://huggingface.co/unsloth/GLM-5.3-GGUF","license":"other (upstream model card); verify terms before commercial use","baseDownloads":712553,"baseLikes":1829,"ggufDownloads":409762,"ggufLikes":77,"capturedAt":"2026-09-14T14:43:34Z"},"terminalBench4":{"rank":9,"agent":"Claude Code","reasoningEffort":"max","trials":330,"confidenceIntervalHalfWidth":3.23,"passAt":[{"attempts":1,"percent":41.82},{"attempts":2,"percent":50.76},{"attempts":3,"percent":54.7},{"attempts":4,"percent":56.67},{"attempts":5,"percent":57.58}]},"fiveBenchmarks":[{"id":"terminal-bench-4","label":"Terminal-Bench","value":41.82,"unit":"percent","version":"4.0","source":"https://www.tbench.ai/","evidence":"official-leaderboard","sourceRecordId":"d72e8775-f4f1-4313-99e7-35b1cb499f24","note":"Claude Code, max effort, 330 trials; ±3.23 percentage points."},{"id":"terminal-bench-3","label":"Terminal-Bench","value":32.4,"unit":"percent","version":"3.0","source":"https://hub.harborframework.com/datasets/terminal-bench/terminal-bench/1?tab=leaderboard&leaderboard=3-0-0","evidence":"official-leaderboard","sourceRecordId":"d5152814-6bf2-473f-be57-b1908ed480a9","note":"Claude Code, max effort; ±2.9 percentage points."},{"id":"terminal-bench-2-1","label":"Terminal-Bench","value":88.2,"unit":"percent","version":"2.1","source":"https://huggingface.co/zai-org/GLM-5.3","evidence":"model-card","sourceRevision":"aca966e4e02791568aa6a4ced368624b3d897f42","note":"Upstream-reported; benchmark versions are not directly comparable."},{"id":"deepswe-1-1","label":"DeepSWE","value":66.9,"unit":"percent","version":"1.1","source":"https://huggingface.co/zai-org/GLM-5.3","evidence":"model-card","sourceRevision":"aca966e4e02791568aa6a4ced368624b3d897f42"},{"id":"swe-marathon-1-1","label":"SWE-Marathon","value":42.5,"unit":"percent","version":"1.1","source":"https://huggingface.co/zai-org/GLM-5.3","evidence":"model-card","sourceRevision":"aca966e4e02791568aa6a4ced368624b3d897f42"}],"artificialAnalysis":{"source":"https://artificialanalysis.ai/models/glm-5-3","relationship":"external-exact-model-reference","observedDimensions":["composite intelligence","hosted output speed","hosted API pricing","context window","model size"],"metricValuesRepublished":false,"proofBoundary":"External comparison reference only. Consult Artificial Analysis directly for current values and terms; hosted observations are not Mac-local performance."},"macFit":{"status":"artifact-fit-estimate-only","smallestArtifact":"UD-IQ1_S (6 shards)","smallestArtifactBytes":216715365893,"smallestArtifactGb":216.7,"lowerLossArtifact":"UD-IQ3_XXS (7 shards)","lowerLossArtifactBytes":281688436357,"lowerLossArtifactGb":281.7,"smallestArtifactFitCandidate":"256GB Mac Studio; 39.3GB remains before runtime, system, and KV cache","lowerLossArtifactFitCandidate":"512GB Mac Studio; 230.3GB remains before runtime, system, and KV cache","calculation":"Decimal unified-memory tier minus total decimal GGUF shard bytes; no runtime overhead or KV-cache reservation is included.","runtimeCandidate":"llama.cpp-compatible GGUF runner; current GLM-5.3 support must be confirmed","measuredOnMac":false,"note":"Unverified artifact-fit estimate, not a run recommendation. Terminal-Bench used a hosted Claude Code configuration, not a Mac-local run."}},{"model":"GLM-5.2","organization":"Z.ai","exactOpenWeightMatch":true,"hf":{"baseRepo":"https://huggingface.co/zai-org/GLM-5.2","ggufRepo":"https://huggingface.co/unsloth/GLM-5.2-GGUF","license":"MIT","baseDownloads":947075,"baseLikes":5111,"ggufDownloads":385343,"ggufLikes":643,"capturedAt":"2026-09-14T14:43:34Z"},"terminalBench4":{"rank":null,"agent":null,"reasoningEffort":null,"trials":null,"confidenceIntervalHalfWidth":null,"passAt":[{"attempts":1,"percent":null},{"attempts":2,"percent":null},{"attempts":3,"percent":null},{"attempts":4,"percent":null},{"attempts":5,"percent":null}]},"fiveBenchmarks":[{"id":"terminal-bench-4","label":"Terminal-Bench","value":null,"unit":"percent","version":"4.0","source":"https://www.tbench.ai/","evidence":"not-listed","note":"No official displayed Terminal-Bench 4.0 entry at capture time."},{"id":"terminal-bench-3","label":"Terminal-Bench","value":4.6,"unit":"percent","version":"3.0","source":"https://hub.harborframework.com/datasets/terminal-bench/terminal-bench/1?tab=leaderboard&leaderboard=3-0-0","evidence":"official-leaderboard","sourceRecordId":"b3c7db0e-0289-4fa5-ac3a-da600b15a72a","note":"Claude Code, max effort; ±1.9 percentage points."},{"id":"terminal-bench-2-1","label":"Terminal-Bench","value":81,"unit":"percent","version":"2.1","source":"https://huggingface.co/zai-org/GLM-5.2","evidence":"model-card","sourceRevision":"cf457fa734ab149ffef225f80893eb38c6ff5cdc","note":"Terminus-2 harness; upstream-reported."},{"id":"deepswe","label":"DeepSWE","value":46.2,"unit":"percent","version":"upstream model card","source":"https://huggingface.co/zai-org/GLM-5.2","evidence":"model-card","sourceRevision":"cf457fa734ab149ffef225f80893eb38c6ff5cdc"},{"id":"swe-marathon","label":"SWE-Marathon","value":13,"unit":"percent","version":"upstream model card","source":"https://huggingface.co/zai-org/GLM-5.2","evidence":"model-card","sourceRevision":"cf457fa734ab149ffef225f80893eb38c6ff5cdc"}],"artificialAnalysis":{"source":"https://artificialanalysis.ai/models/glm-5-2","relationship":"external-exact-model-reference","observedDimensions":["composite intelligence","hosted output speed","hosted API pricing","context window","model size"],"metricValuesRepublished":false,"proofBoundary":"External comparison reference only. Consult Artificial Analysis directly for current values and terms; hosted observations are not Mac-local performance."},"macFit":{"status":"artifact-fit-estimate-only","smallestArtifact":"UD-IQ1_S (6 shards)","smallestArtifactBytes":216715360960,"smallestArtifactGb":216.7,"lowerLossArtifact":"UD-IQ3_XXS (7 shards)","lowerLossArtifactBytes":281688431424,"lowerLossArtifactGb":281.7,"smallestArtifactFitCandidate":"256GB Mac Studio; 39.3GB remains before runtime, system, and KV cache","lowerLossArtifactFitCandidate":"512GB Mac Studio; 230.3GB remains before runtime, system, and KV cache","calculation":"Decimal unified-memory tier minus total decimal GGUF shard bytes; no runtime overhead or KV-cache reservation is included.","runtimeCandidate":"llama.cpp-compatible GGUF runner; current GLM-5.2 support must be confirmed","measuredOnMac":false,"note":"Unverified artifact-fit estimate, not a run recommendation. The leaderboard score belongs to the displayed model-agent setup, not every quantization."}}],"provenance":{"evidenceSnapshot":"data/evidence/terminal-bench-mac-index-2026-09-14.json","evidenceSnapshotSha256":"f3352b0f4beccd934b26e8f6af2cbfac8a3d7c7e39033ffc14f05fe95acdb47f","terminalBench4LeaderboardId":"9f966760-00f1-424e-90f5-c964fb6f6091","terminalBench3LeaderboardId":"694bb3b0-c5d5-4899-93de-4c392b814d7e","hfBaseRevisions":{"GLM-5.3":"aca966e4e02791568aa6a4ced368624b3d897f42","GLM-5.2":"cf457fa734ab149ffef225f80893eb38c6ff5cdc"},"hfGgufRevisions":{"GLM-5.3":"346b3591c7f28d1a23716f97a065ecf12ec14771","GLM-5.2":"abc55e72527792c6e77069c99b4cb7de16fa9f23"}},"limitations":["Only exact open-weight models displayed on official Terminal-Bench 3.0 or 4.0 leaderboards are included.","Terminal-Bench results measure a model plus agent harness and evaluation configuration; they do not measure Mac throughput.","Quantization can change capability, and the listed GGUF artifacts were not rerun on Terminal-Bench, launched on a Mac, or measured on a Mac by MacLocalModels.","Mac tiers are arithmetic artifact-fit candidates only: unified-memory tier minus GGUF shard bytes, before runtime, operating-system, and KV-cache memory.","Pass@2 through pass@5 are retry-success estimates from Terminal-Bench 4.0, not five independent benchmark suites.","Hugging Face downloads and likes are point-in-time signals, not quality scores.","Artificial Analysis is linked as an external exact-model comparison reference; its metric values are not republished in this dataset.","Artificial Analysis hosted-provider observations are not local Apple Silicon inference and should not be combined with Terminal-Bench as independent evidence without reviewing its current methodology."],"sources":{"leaderboard4":"https://www.tbench.ai/","leaderboard3":"https://hub.harborframework.com/datasets/terminal-bench/terminal-bench/1?tab=leaderboard&leaderboard=3-0-0","dataset":"https://github.com/harbor-framework/terminal-bench","glm53":"https://huggingface.co/zai-org/GLM-5.3","glm53Gguf":"https://huggingface.co/unsloth/GLM-5.3-GGUF","glm52":"https://huggingface.co/zai-org/GLM-5.2","glm52Gguf":"https://huggingface.co/unsloth/GLM-5.2-GGUF","artificialAnalysisIndex":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index","artificialAnalysisMethodology":"https://artificialanalysis.ai/methodology/intelligence-benchmarking","artificialAnalysisGlm53":"https://artificialanalysis.ai/models/glm-5-3","artificialAnalysisGlm52":"https://artificialanalysis.ai/models/glm-5-2"}}