📐 计分规则Scoring Rules
能力上限 = 加权原始分 = ToolCall×0.3 + BugFind×0.3 + HermesAgent×0.4。反映模型的最高能力水平。Max Score = Weighted raw score = ToolCall×0.3 + BugFind×0.3 + HermesAgent×0.4. Reflects the model's highest capability level.
实用得分 = 能力上限 − 重试扣分。每题首次通过不扣分;重试后才通过的题目每重试1次扣1分;重试后仍失败的不参与扣分。Effective Score = Max Score − retry penalty. First pass = no penalty; each retry before passing = −1pt; failed retries excluded.
本模型能力上限 93.2,实用得分 85.2(扣分 8 分)。重试代价远低于思考版(18 分),实用性突出。This model: Max Score 93.2, Effective Score 85.2 (−8 penalty). Retry cost much lower than thinking version (18), highlighting practical reliability.
| 题号ID | 分数Score | 结果Result | 重试Retry | 说明Summary |
|---|
| 题号ID | 分数Score | 结果Result | 重试Retry | 说明Summary |
|---|
| 题号ID | 分数Score | 结果Result | 重试Retry | 说明Summary |
|---|
| 题号ID | 分数Score | 结果Result | 重试Retry | 失败原因Failure Reason |
|---|
📊 模型评估总结Model Evaluation Summary
Ornith-1.0-35B-MTP(无思考模式)展现了扎实且稳定的实力。ToolCall-15 满分通过(仅 TC-03 首次 partial 后重试通过),BugFind-15 能力上限 92 分且通过了红鲱鱼陷阱题(BF-10 满分),HermesAgent-20 通过率 85%(17/20),远优于思考版的 75%。Ornith-1.0-35B-MTP (No Thinking) demonstrates solid and stable performance. ToolCall-15 perfect score (only TC-03 first attempt partial, passed on retry), BugFind-15 ceiling 92 and passed the red herring trap (BF-10 full marks), HermesAgent-20 85% pass rate (17/20), far better than the thinking version's 75%.
无思考模式最大的优势在于稳定性:总重试 8 次仅扣 8 分(思考版 18 次扣 18 分)。特别是 HA-17(1 次重试后通过 vs 思考版失败 7 次仍没通过)和 BF-10(100 分 vs 思考版 0 分)的表现差距明显。弱点是 BF-03(与思考版一样踩坑)、BF-09(slice aliasing 诊断不完整,3 次重试仍 partial)、HA-16/19/20(5 次重试仍失败)。The no-thinking mode's biggest advantage is stability: 8 retries cost only 8 points (thinking version: 18 retries, 18 points). Especially HA-17 (passed after 1 retry vs thinking version's 7 retries still failing) and BF-10 (100 vs thinking 0) show clear gaps. Weaknesses: BF-03 (same trap as thinking), BF-09 (incomplete slice aliasing diagnosis, 3 retries still partial), and HA-16/19/20 (5 retries all failed).
能力上限 93.2(全场第三),实用得分 85.2——不思考反而更靠谱的典范。Ceiling 93.2 (rank #3), effective 85.2 — a textbook case of "not thinking" being more reliable.