不检测渠道接口规范, 只检测渠道任务能力, 通过多种渠道统计渠道差异
数据更新于 · 26 天前
测试次数
1,888
测试题目
21
总计花费
$1234.45
| 站点 | 渠道 | 模型 | 通过率 | 平均花费 (USD) | Token TPS | 测试次数 |
|---|---|---|---|---|---|---|
| Micu | OFFICIAL | gpt-6-astra | 100% | $1.2806 | 7.1 | 2 |
| CCTQ | OFFICIAL | gpt-6-astra | 100% | $0.4006 | 12.9 | 2 |
| ByteCat | OFFICIAL | gpt-6-astra | 100% | $0.6694 | 9.3 | 2 |
| IKun Code | OFFICIAL | gpt-6-astra | 100% | $1.5255 | 17.4 | 2 |
| 4Router | OFFICIAL | claude-opus-5 | 100% | $1.6877 | 61.6 | 2 |
| Neko Code | OFFICIAL | gpt-6-astra | 100% | $0.7187 | 14.9 | 2 |
| TimiCC | OFFICIAL | gpt-6-astra | 100% | $0.7563 | 16.1 | 2 |
| Yes Code | OFFICIAL | gpt-6-astra | 100% | $0.9662 | 17.4 | 2 |
| 4Router | OFFICIAL | gpt-6-astra | 100% | $0.3936 | 15.6 | 2 |
| Duck Code | OFFICIAL | gpt-5.6-sol | 80% | $0.4783 | 27.9 | 15 |
| CCTQ | OFFICIAL | claude-opus-5 | 70.6% | $1.4250 | 56.6 | 17 |
| Packy Code | OFFICIAL | claude-opus-5 | 70.6% | $2.4525 | 50.7 | 17 |
| Micu | OFFICIAL | claude-opus-5 | 70.6% | $1.6934 | 57.2 | 17 |
| ByteCat | OFFICIAL | claude-opus-5 | 70.6% | $1.2541 | 56.0 | 17 |
| Yunwu | OFFICIAL | claude-opus-5 | 66.7% | $0.5225 | 27.5 | 12 |
| APINebula | OFFICIAL | claude-opus-5 | 64.7% | $2.0490 | 54.9 | 17 |
| 78 Code | OFFICIAL | claude-opus-5 | 64.7% | $0.7006 | 57.9 | 17 |
| Right Code | OFFICIAL | claude-opus-5 | 64.7% | $1.7648 | 51.0 | 17 |
| IKun Code | OFFICIAL | claude-opus-5 | 64.7% | $3.5009 | 51.5 | 17 |
| SSSAI Code | OFFICIAL | claude-opus-5 | 58.8% | $1.2603 | 56.0 | 17 |
| Neko Code | OFFICIAL | claude-opus-5 | 58.8% | $1.3155 | 69.7 | 17 |
| TimiCC | OFFICIAL | claude-opus-5 | 58.8% | $1.0660 | 52.6 | 17 |
| Duck Code | OFFICIAL | claude-opus-5 | 53.3% | $1.2128 | 53.1 | 15 |
| Yes Code | OFFICIAL | claude-opus-5 | 52.9% | $1.1311 | 48.2 | 17 |
| Somebody | AWSQ | claude-opus-5 | 52.9% | $1.2041* | 41.2 | 17 |
| Packy Code | OFFICIAL | gpt-6-astra | 50% | $1.4193 | 12.3 | 2 |
| SSSAI Code | OFFICIAL | gpt-6-astra | 50% | $0.1167 | 6.2 | 2 |
| APINebula | OFFICIAL | gpt-6-astra | 50% | $0.4550 | 10.8 | 2 |
| 78 Code | OFFICIAL | gpt-6-astra | 50% | $0.2008 | 8.5 | 2 |
| Right Code | OFFICIAL | gpt-6-astra | 50% | $1.0041 | 10.3 | 2 |
| Cubence | OFFICIAL | claude-opus-5 | 47.1% | $1.3276 | 44.6 | 17 |
| Cubence | OFFICIAL | gpt-6-astra | 0% | — | — | 2 |
花费为模型官方牌价 × 站点最近已知折扣的折算值；带 * 的行未匹配到站点折扣，按官方牌价估算。
按全部历史 trial 聚合的渠道排名，口径见下方评测方法论。
| # | 渠道 | 模型 | 通过率 | 速度 | 成本 |
|---|---|---|---|---|---|
| 1 | Micu · OFFICIAL | gpt-6-astra | 100.0% | 7.1 | $1.2806 |
| 2 | CCTQ · OFFICIAL | gpt-6-astra | 100.0% | 12.9 | $0.4006 |
| 3 | ByteCat · OFFICIAL | gpt-6-astra | 100.0% | 9.3 | $0.6694 |
| 4 | IKun Code · OFFICIAL | gpt-6-astra | 100.0% | 17.4 | $1.5255 |
| 5 | 4Router · OFFICIAL | claude-opus-5 | 100.0% | 61.6 | $1.6877 |
| 6 | Neko Code · OFFICIAL | gpt-6-astra | 100.0% | 14.9 | $0.7187 |
| 7 | TimiCC · OFFICIAL | gpt-6-astra | 100.0% | 16.1 | $0.7563 |
| 8 | Yes Code · OFFICIAL | gpt-6-astra | 100.0% | 17.4 | $0.9662 |
| 9 | 4Router · OFFICIAL | gpt-6-astra | 100.0% | 15.6 | $0.3936 |
| 10 | Duck Code · OFFICIAL | gpt-5.6-sol | 80.0% | 27.9 | $0.4783 |
| # | 渠道 | 模型 | 通过率 | 速度 | 成本 |
|---|---|---|---|---|---|
| 1 | Neko Code · OFFICIAL | claude-opus-5 | 58.8% | 69.7 | $1.3155 |
| 2 | 4Router · OFFICIAL | claude-opus-5 | 100.0% | 61.6 | $1.6877 |
| 3 | 78 Code · OFFICIAL | claude-opus-5 | 64.7% | 57.9 | $0.7006 |
| 4 | Micu · OFFICIAL | claude-opus-5 | 70.6% | 57.2 | $1.6934 |
| 5 | CCTQ · OFFICIAL | claude-opus-5 | 70.6% | 56.6 | $1.4250 |
| # | 渠道 | 模型 | 通过率 | 速度 | 成本 |
|---|---|---|---|---|---|
| 1 | SSSAI Code · OFFICIAL | gpt-6-astra | 50.0% | 6.2 | $0.1167 |
| 2 | 78 Code · OFFICIAL | gpt-6-astra | 50.0% | 8.5 | $0.2008 |
| 3 | 4Router · OFFICIAL | gpt-6-astra | 100.0% | 15.6 | $0.3936 |
| 4 | CCTQ · OFFICIAL | gpt-6-astra | 100.0% | 12.9 | $0.4006 |
| 5 | APINebula · OFFICIAL | gpt-6-astra | 50.0% | 10.8 | $0.4550 |
| 6 |
截至 2026年9月8日，本站已对 32 个渠道配置在 21 道真实编程任务上累计执行 1888 次测试，总花费约 $1234，全部自费、无中转赞助。
以测试题为主键，展示在 terminal-bench / aider-polyglot 等真实编程任务下的跑分，而非接口连通性检测。
每次 trial 在隔离容器内运行：CLI Agent（codex / claude-code 等）通过被测渠道调用模型完成任务，随后由独立 Verifier 执行任务自带的验收脚本判分，全程记录 Token 消耗、耗时与实际花费。
仅 Verifier 验收通过的 trial 计为通过；模型未完成任务、运行异常或超时一律计为失败，不做人工补测，也不挑选成绩重跑。
历史通过率 = 通过 trial 数 ÷ 总 trial 数 × 100%；水平线 = 所有 config 历史通过率的算术平均。
统计每个渠道（config）的全部历史 trial；卡片明细基于最新一次已上架 run 的切片，历史通过率跨全部已上架 run 累计。同 task/config 历史 trial 数少于 3 次标记「样本不足」，避免小样本误导。指标含通过率、平均 Token、平均花费、Token TPS。
| 6 |
| ByteCat · OFFICIAL |
| claude-opus-5 |
| 70.6% |
| 56.0 |
| $1.2541 |
| 7 | SSSAI Code · OFFICIAL | claude-opus-5 | 58.8% | 56.0 | $1.2603 |
| 8 | APINebula · OFFICIAL | claude-opus-5 | 64.7% | 54.9 | $2.0490 |
| 9 | Duck Code · OFFICIAL | claude-opus-5 | 53.3% | 53.1 | $1.2128 |
| 10 | TimiCC · OFFICIAL | claude-opus-5 | 58.8% | 52.6 | $1.0660 |
| Duck Code · OFFICIAL |
| gpt-5.6-sol |
| 80.0% |
| 27.9 |
| $0.4783 |
| 7 | Yunwu · OFFICIAL | claude-opus-5 | 66.7% | 27.5 | $0.5225 |
| 8 | ByteCat · OFFICIAL | gpt-6-astra | 100.0% | 9.3 | $0.6694 |
| 9 | 78 Code · OFFICIAL | claude-opus-5 | 64.7% | 57.9 | $0.7006 |
| 10 | Neko Code · OFFICIAL | gpt-6-astra | 100.0% | 14.9 | $0.7187 |
