Evaluation results under 128K context length. The card recommends temperature=1.0 and, for code-related tasks such as LCB, top_p=0.95 and top_k=40. Per-benchmark scaffold, sample count, metric and HLE full/text-only subset are not stated.
条件
GLM-4.6; chart column without w/Tools label; task version/protocol unspecified except the benchmark label. No unit inferred where the source does not print one. 2026-11-18按完整模型卡修正单位、指标/题集或协议标注,数字不变。
GLM-4.6 · SWE-bench Verified · GLM-4.6模型卡;Verified · Scores and configurations
GLM-4.6 record for SWE-bench Verified · GLM-4.6模型卡;Verified. 1 score record for this release, including 1 exact value (may include historical records). Includes 2 source locations.
Evaluation results under 128K context length. The card recommends temperature=1.0 and, for code-related tasks such as LCB, top_p=0.95 and top_k=40. Per-benchmark scaffold, sample count, metric and HLE full/text-only subset are not stated.
Conditions
GLM-4.6; chart column without w/Tools label; task version/protocol unspecified except the benchmark label. No unit inferred where the source does not print one. 2026-11-18按完整模型卡修正单位、指标/题集或协议标注,数字不变。