← 返回 API Dashboard

本机回环 API · 本页为 registry 静态快照

Free Model Chain — 模型分数全览

展示全部免费模型的分层评分与 routing_score 加权计算过程。数据来自 model-registry-latest.json，非实时调用 8022 API。API 仍仅本机 127.0.0.1:8022 可访问。

更新时间
2026-06-06 01:30:33 CST
scoring_version
v2-routing
quality 覆盖
19 / 27 模型已 quality 评测（0.7037）
数据来源
reports/run-20260606-011232/benchmark.json + quality benchmarks merged + newer partial gate runs merged

general 推荐链（按 routing 排序）

  1. openrouter/owl-alpha
  2. google/gemma-4-31b-it:free
  3. openai/gpt-oss-120b:free
  4. poolside/laguna-xs.2:free
  5. nvidia/nemotron-3-super-120b-a12b:free
  6. nvidia/nemotron-nano-12b-v2-vl:free

buildsignal 推荐链

  1. openrouter/owl-alpha
  2. google/gemma-4-31b-it:free
  3. openai/gpt-oss-120b:free
  4. poolside/laguna-xs.2:free
  5. nvidia/nemotron-3-super-120b-a12b:free
  6. nvidia/nemotron-nano-12b-v2-vl:free
#1

openrouter/owl-alpha

0.9575 routing_score 推荐

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.9667 ×0.45 0.4350
gate_score 1.0000 ×0.20 0.2000
stability_score 0.9000 ×0.20 0.1800
format_score 1.0000 ×0.10 0.1000
latency_score (9630 ms) 0.8500 ×0.05 0.0425
合计 → routing_score 0.9575

② Gate 层（smoke test）

gate_score = 1.0000（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning1.0000
BuildSignal 片段1.0000

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.9667
effective_quality_score0.9667
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 1.0000
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 9630
  • BuildSignal recommended: 推荐
#2

google/gemma-4-31b-it:free

0.9540 routing_score 推荐

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.9867 ×0.45 0.4440
gate_score 0.9375 ×0.20 0.1875
stability_score 0.9000 ×0.20 0.1800
format_score 1.0000 ×0.10 0.1000
latency_score (8363 ms) 0.8500 ×0.05 0.0425
合计 → routing_score 0.9540

② Gate 层（smoke test）

gate_score = 0.9375（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning1.0000
BuildSignal 片段0.7500

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.9867
effective_quality_score0.9867
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 1.0000
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 8363
  • BuildSignal recommended: 推荐
#3

openai/gpt-oss-120b:free

0.9425 routing_score 推荐

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.9600 ×0.45 0.4320
gate_score 1.0000 ×0.20 0.2000
stability_score 0.9000 ×0.20 0.1800
format_score 0.9800 ×0.10 0.0980
latency_score (14695 ms) 0.6500 ×0.05 0.0325
合计 → routing_score 0.9425

② Gate 层（smoke test）

gate_score = 1.0000（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning1.0000
BuildSignal 片段1.0000

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.9600
effective_quality_score0.9600
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.9000
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 14695
  • BuildSignal recommended: 推荐
#4

poolside/laguna-xs.2:free

0.9360 routing_score 推荐

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.9467 ×0.45 0.4260
gate_score 0.9375 ×0.20 0.1875
stability_score 0.9000 ×0.20 0.1800
format_score 1.0000 ×0.10 0.1000
latency_score (8086 ms) 0.8500 ×0.05 0.0425
合计 → routing_score 0.9360

② Gate 层（smoke test）

gate_score = 0.9375（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning1.0000
BuildSignal 片段0.7500

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.9467
effective_quality_score0.9467
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 1.0000
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 8086
  • BuildSignal recommended: 推荐
#5

nvidia/nemotron-3-super-120b-a12b:free

0.8416 routing_score 推荐

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.8063 ×0.45 0.3628
gate_score 1.0000 ×0.20 0.2000
stability_score 0.9000 ×0.20 0.1800
format_score 0.7880 ×0.10 0.0788
latency_score (47176 ms) 0.4000 ×0.05 0.0200
合计 → routing_score 0.8416

② Gate 层（smoke test）

gate_score = 1.0000（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning1.0000
BuildSignal 片段1.0000

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.8063
effective_quality_score0.8063
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.4400
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 47176
  • BuildSignal recommended: 推荐
#6

nvidia/nemotron-nano-12b-v2-vl:free

0.7945 routing_score 推荐

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.6933 ×0.45 0.3120
gate_score 0.9375 ×0.20 0.1875
stability_score 0.9000 ×0.20 0.1800
format_score 0.8250 ×0.10 0.0825
latency_score (12061 ms) 0.6500 ×0.05 0.0325
合计 → routing_score 0.7945

② Gate 层（smoke test）

gate_score = 0.9375（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning1.0000
BuildSignal 片段0.7500

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.6933
effective_quality_score0.6933
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 1.0000
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 12061
  • BuildSignal recommended: 推荐
#7

liquid/lfm-2.5-1.2b-instruct:free

0.7656 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.8287 ×0.45 0.3729
gate_score 0.6250 ×0.20 0.1250
stability_score 0.7000 ×0.20 0.1400
format_score 0.8520 ×0.10 0.0852
latency_score (5473 ms) 0.8500 ×0.05 0.0425
合计 → routing_score 0.7656

② Gate 层（smoke test）

gate_score = 0.6250（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning0.5000
BuildSignal 片段0.0000

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.8287
effective_quality_score0.8287
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.2600
reasoning_score (legacy)0.5000 = gate 推理

④ 其它

  • latency_ms: 5473
  • BuildSignal recommended: —
#8

nvidia/nemotron-3-nano-30b-a3b:free

0.7116 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.6253 ×0.45 0.2814
gate_score 0.8125 ×0.20 0.1625
stability_score 0.7000 ×0.20 0.1400
format_score 0.8520 ×0.10 0.0852
latency_score (8169 ms) 0.8500 ×0.05 0.0425
合计 → routing_score 0.7116

② Gate 层（smoke test）

gate_score = 0.8125（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning1.0000
BuildSignal 片段0.2500

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.6253
effective_quality_score0.6253
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.2600
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 8169
  • BuildSignal recommended: —
#9

z-ai/glm-4.5-air:free

0.6916 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.6253 ×0.45 0.2814
gate_score 0.8750 ×0.20 0.1750
stability_score 0.7000 ×0.20 0.1400
format_score 0.7520 ×0.10 0.0752
latency_score (33213 ms) 0.4000 ×0.05 0.0200
合计 → routing_score 0.6916

② Gate 层（smoke test）

gate_score = 0.8750（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning1.0000
BuildSignal 片段0.5000

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.6253
effective_quality_score0.6253
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.2600
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 33213
  • BuildSignal recommended: —
#10

nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free

0.6893 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.6203 ×0.45 0.2791
gate_score 0.8125 ×0.20 0.1625
stability_score 0.7000 ×0.20 0.1400
format_score 0.6520 ×0.10 0.0652
latency_score (7626 ms) 0.8500 ×0.05 0.0425
合计 → routing_score 0.6893

② Gate 层（smoke test）

gate_score = 0.8125（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning1.0000
BuildSignal 片段0.2500

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.6203
effective_quality_score0.6203
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.2600
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 7626
  • BuildSignal recommended: —
#11

openrouter/free

0.6826 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.6387 ×0.45 0.2874
gate_score 0.6875 ×0.20 0.1375
stability_score 0.7000 ×0.20 0.1400
format_score 0.8520 ×0.10 0.0852
latency_score (14288 ms) 0.6500 ×0.05 0.0325
合计 → routing_score 0.6826

② Gate 层（smoke test）

gate_score = 0.6875（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning0.0000
BuildSignal 片段0.7500

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.6387
effective_quality_score0.6387
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.2600
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 14288
  • BuildSignal recommended: —
#12

poolside/laguna-m.1:free

0.6701 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.6387 ×0.45 0.2874
gate_score 0.6875 ×0.20 0.1375
stability_score 0.7000 ×0.20 0.1400
format_score 0.8520 ×0.10 0.0852
latency_score (69297 ms) 0.4000 ×0.05 0.0200
合计 → routing_score 0.6701

② Gate 层（smoke test）

gate_score = 0.6875（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning0.0000
BuildSignal 片段0.7500

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.6387
effective_quality_score0.6387
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.2600
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 69297
  • BuildSignal recommended: —
#13

nvidia/nemotron-3-ultra-550b-a55b:free

0.6394 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.4680 ×0.45 0.2106
gate_score 0.9375 ×0.20 0.1875
stability_score 0.7000 ×0.20 0.1400
format_score 0.8133 ×0.10 0.0813
latency_score (79850 ms) 0.4000 ×0.05 0.0200
合计 → routing_score 0.6394

② Gate 层（smoke test）

gate_score = 0.9375（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning1.0000
BuildSignal 片段0.7500

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.4680
effective_quality_score0.4680
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.4400
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 79850
  • BuildSignal recommended: —
#14

openai/gpt-oss-20b:free

0.5700 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.3667 ×0.45 0.1650
gate_score 0.6875 ×0.20 0.1375
stability_score 0.7000 ×0.20 0.1400
format_score 0.9500 ×0.10 0.0950
latency_score (13910 ms) 0.6500 ×0.05 0.0325
合计 → routing_score 0.5700

② Gate 层（smoke test）

gate_score = 0.6875（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct0.0000
简单推理 reasoning1.0000
BuildSignal 片段0.7500

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.3667
effective_quality_score0.3667
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.9000
reasoning_score (legacy)1.0000 = gate 推理

④ 其它

  • latency_ms: 13910
  • BuildSignal recommended: —
#15

nvidia/nemotron-nano-9b-v2:free

0.5676 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.4887 ×0.45 0.2199
gate_score 0.5000 ×0.20 0.1000
stability_score 0.7000 ×0.20 0.1400
format_score 0.7520 ×0.10 0.0752
latency_score (17797 ms) 0.6500 ×0.05 0.0325
合计 → routing_score 0.5676

② Gate 层（smoke test）

gate_score = 0.5000（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.4887
effective_quality_score0.4887
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.2600
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 17797
  • BuildSignal recommended: —
#16

liquid/lfm-2.5-1.2b-thinking:free

0.5626 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.4553 ×0.45 0.2049
gate_score 0.5000 ×0.20 0.1000
stability_score 0.7000 ×0.20 0.1400
format_score 0.8520 ×0.10 0.0852
latency_score (13601 ms) 0.6500 ×0.05 0.0325
合计 → routing_score 0.5626

② Gate 层（smoke test）

gate_score = 0.5000（四题通过率均值）

中文 zh_summary1.0000
JSON json_struct1.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.4553
effective_quality_score0.4553
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.2600
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 13601
  • BuildSignal recommended: —
#17

google/gemma-4-26b-a4b-it:free

0.3665 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.1867 ×0.45 0.0840
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 1.0000 ×0.10 0.1000
latency_score (5146 ms) 0.8500 ×0.05 0.0425
合计 → routing_score 0.3665

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.1867
effective_quality_score0.1867
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.0000
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 5146
  • BuildSignal recommended: —
#18

moonshotai/kimi-k2.6:free

0.3505 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.1733 ×0.45 0.0780
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 1.0000 ×0.10 0.1000
latency_score (17390 ms) 0.6500 ×0.05 0.0325
合计 → routing_score 0.3505

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.1733
effective_quality_score0.1733
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.0000
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 17390
  • BuildSignal recommended: —
#19

nvidia/nemotron-3.5-content-safety:free

0.3398 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.2270 ×0.45 0.1022
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 0.5520 ×0.10 0.0552
latency_score (6491 ms) 0.8500 ×0.05 0.0425
合计 → routing_score 0.3399

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

quality_score 来自 5 道 quality 题平均分（见仓库 data/README.md）。

quality_score0.2270
effective_quality_score0.2270
quality_score_sourcequality_benchmark
quality_penalty0.0000
quality_evaluated是
quality_reasoning（多约束推理）: 0.2600
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 6491
  • BuildSignal recommended: —
#20

cognitivecomputations/dolphin-mistral-24b-venice-edition:free

0.1400 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.0000 ×0.45 0.0000
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 0.0000 ×0.10 0.0000
latency_score (0 ms) 0.0000 ×0.05 0.0000
合计 → routing_score 0.1400

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

未跑 quality benchmark：quality_score 为 gate 兜底；effective = gate − penalty (0.0000 − 0.1000)。quality_score_missing; used gate_score as fallback

quality_score0.0000
effective_quality_score0.0000
quality_score_sourcegate_fallback
quality_penalty0.1000
quality_evaluated否
quality_reasoning: 未评测
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 0
  • BuildSignal recommended: —
#21

google/lyria-3-clip-preview

0.1400 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.0000 ×0.45 0.0000
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 0.0000 ×0.10 0.0000
latency_score (0 ms) 0.0000 ×0.05 0.0000
合计 → routing_score 0.1400

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

未跑 quality benchmark：quality_score 为 gate 兜底；effective = gate − penalty (0.0000 − 0.1000)。quality_score_missing; used gate_score as fallback

quality_score0.0000
effective_quality_score0.0000
quality_score_sourcegate_fallback
quality_penalty0.1000
quality_evaluated否
quality_reasoning: 未评测
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 0
  • BuildSignal recommended: —
#22

google/lyria-3-pro-preview

0.1400 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.0000 ×0.45 0.0000
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 0.0000 ×0.10 0.0000
latency_score (0 ms) 0.0000 ×0.05 0.0000
合计 → routing_score 0.1400

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

未跑 quality benchmark：quality_score 为 gate 兜底；effective = gate − penalty (0.0000 − 0.1000)。quality_score_missing; used gate_score as fallback

quality_score0.0000
effective_quality_score0.0000
quality_score_sourcegate_fallback
quality_penalty0.1000
quality_evaluated否
quality_reasoning: 未评测
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 0
  • BuildSignal recommended: —
#23

meta-llama/llama-3.2-3b-instruct:free

0.1400 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.0000 ×0.45 0.0000
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 0.0000 ×0.10 0.0000
latency_score (0 ms) 0.0000 ×0.05 0.0000
合计 → routing_score 0.1400

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

未跑 quality benchmark：quality_score 为 gate 兜底；effective = gate − penalty (0.0000 − 0.1000)。quality_score_missing; used gate_score as fallback

quality_score0.0000
effective_quality_score0.0000
quality_score_sourcegate_fallback
quality_penalty0.1000
quality_evaluated否
quality_reasoning: 未评测
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 0
  • BuildSignal recommended: —
#24

meta-llama/llama-3.3-70b-instruct:free

0.1400 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.0000 ×0.45 0.0000
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 0.0000 ×0.10 0.0000
latency_score (0 ms) 0.0000 ×0.05 0.0000
合计 → routing_score 0.1400

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

未跑 quality benchmark：quality_score 为 gate 兜底；effective = gate − penalty (0.0000 − 0.1000)。quality_score_missing; used gate_score as fallback

quality_score0.0000
effective_quality_score0.0000
quality_score_sourcegate_fallback
quality_penalty0.1000
quality_evaluated否
quality_reasoning: 未评测
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 0
  • BuildSignal recommended: —
#25

nousresearch/hermes-3-llama-3.1-405b:free

0.1400 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.0000 ×0.45 0.0000
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 0.0000 ×0.10 0.0000
latency_score (0 ms) 0.0000 ×0.05 0.0000
合计 → routing_score 0.1400

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

未跑 quality benchmark：quality_score 为 gate 兜底；effective = gate − penalty (0.0000 − 0.1000)。quality_score_missing; used gate_score as fallback

quality_score0.0000
effective_quality_score0.0000
quality_score_sourcegate_fallback
quality_penalty0.1000
quality_evaluated否
quality_reasoning: 未评测
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 0
  • BuildSignal recommended: —
#26

qwen/qwen3-coder:free

0.1400 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.0000 ×0.45 0.0000
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 0.0000 ×0.10 0.0000
latency_score (0 ms) 0.0000 ×0.05 0.0000
合计 → routing_score 0.1400

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

未跑 quality benchmark：quality_score 为 gate 兜底；effective = gate − penalty (0.0000 − 0.1000)。quality_score_missing; used gate_score as fallback

quality_score0.0000
effective_quality_score0.0000
quality_score_sourcegate_fallback
quality_penalty0.1000
quality_evaluated否
quality_reasoning: 未评测
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 0
  • BuildSignal recommended: —
#27

qwen/qwen3-next-80b-a3b-instruct:free

0.1400 routing_score —

① 最终分 routing_score 计算

routing = effective_quality×0.45 + gate×0.20 + stability×0.20 + format×0.10 + latency_score×0.05

项原始分×权重加权贡献
effective_quality_score 0.0000 ×0.45 0.0000
gate_score 0.0000 ×0.20 0.0000
stability_score 0.7000 ×0.20 0.1400
format_score 0.0000 ×0.10 0.0000
latency_score (0 ms) 0.0000 ×0.05 0.0000
合计 → routing_score 0.1400

② Gate 层（smoke test）

gate_score = 0.0000（四题通过率均值）

中文 zh_summary0.0000
JSON json_struct0.0000
简单推理 reasoning0.0000
BuildSignal 片段0.0000

③ Quality 层

未跑 quality benchmark：quality_score 为 gate 兜底；effective = gate − penalty (0.0000 − 0.1000)。quality_score_missing; used gate_score as fallback

quality_score0.0000
effective_quality_score0.0000
quality_score_sourcegate_fallback
quality_penalty0.1000
quality_evaluated否
quality_reasoning: 未评测
reasoning_score (legacy)0.0000 = gate 推理

④ 其它

  • latency_ms: 0
  • BuildSignal recommended: —

评分细则见仓库 00-projects/22-openrouter-free-benchmark/data/README.md。本页随 build_model_registry.py 自动刷新。