KAT-Coder-Air V2.5 (high) leads on average score with 5.6 vs 5.5. Qwen3.8 27B has the lower benchmark cost at ~$0.006 vs $0.077. Qwen3.8 27B is faster at 3.12s vs 15.87s, with pass rates of 42.4% vs 40.9%.
Last updated at:
2026-09-28
Compared models
Rank
#265
Total Output Tokens
114,654
Response Time (avg)
15.87s
Total Cost
$0.077
Candidate responses were generated on local hardware. OpenRouter was used only for judging.
Rank
#277
Total Output Tokens
11,445
Response Time (avg)
3.12s
Total Cost
~$0.006Estimated GPU electricity cost only. It excludes the rest of the computer and hardware ownership. NVIDIA GeForce RTX 3090: 0.0572 h × 0.300 kW × €0.2896/kWh × 1.138 USD/EUR = ~$0.005659.
Recommended modelQwen3.8 27B
Its score stays close to the best score here (5.5 vs 5.6), while costing about 13.5x less than KAT-Coder-Air V2.5 (high).
Qwen3.8 27BQwen3.8 27BnoneCandidate responses were generated on local hardware. OpenRouter was used only for judging.Release: 2026-08-14
Score
5.6Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
5.5Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
Rank
#265
#277
Reliability
9.8First-attempt success score: 10.0 means no retryable target API or rate-limit failures before successful calls; tracked failures lower the score.…
10.0First-attempt success score: 10.0 means no retryable target API or rate-limit failures before successful calls; tracked failures lower the score.…
Consistency
8.1Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
Attempts
66/66
66/66
Tests Correct
A test is fully passed only if every run passed for that test.Wrong answer: 9API error: 3Extra formatting: 3Response Time (avg)15.87sResponse Time (max)118.35sResponse Time (total)349.16sA test is fully passed only if every run passed for that test.…
A test is fully passed only if every run passed for that test.Wrong answer: 11Did not follow instructions: 1Invalid tool call: 1Response Time (avg)3.12sResponse Time (max)44.76sResponse Time (total)68.69sA test is fully passed only if every run passed for that test.…
Attempt pass rate
42.4%Attempt pass rate = passed attempts / total attempts across runs.…
40.9%Attempt pass rate = passed attempts / total attempts across runs.…
Flaky tests
5Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
Total Runs
66Total Runs…
66Total Runs…
Cost per result
1.091
~0.063Estimated GPU electricity cost only. It excludes the rest of the computer and hardware ownership. NVIDIA GeForce RTX 3090: 0.0572 h × 0.300 kW × €0.2896/kWh × 1.138 USD/EUR = ~$0.005659.
Total Cost
$0.077
~$0.006Estimated GPU electricity cost only. It excludes the rest of the computer and hardware ownership. NVIDIA GeForce RTX 3090: 0.0572 h × 0.300 kW × €0.2896/kWh × 1.138 USD/EUR = ~$0.005659.
Input Price
$0.150 / 1MInput Price…
N/AInput Price…
Output Price
$0.600 / 1MOutput Price…
N/AOutput Price…
Total Input Tokens
50,423Total Input Tokens…
122,463Total Input Tokens…
Output Tokens
4,144Output Tokens…
11,445Output Tokens…
Reasoning Tokens
110,510Reasoning Tokens…
0Reasoning Tokens…
Response Time (avg)
15.87sResponse Time (avg)…
3.12sResponse Time (avg)…
Response Time (max)
118.35sResponse Time (max)…
44.76sResponse Time (max)…
Response Time (total)
349.16sResponse Time (total)…
68.69sResponse Time (total)…
Parameters
~35B total (~3B active)
27.3B
Availability
Closed
Weights available
Model generation showcase
Hamster playing table tennis
Prompt: Create a detailed SVG illustration of a hamster playing table tennis.
#265 KAT-Coder-Air V2.5
high
Reached the allocated time limit (300 seconds) without receiving showcase output.
Cost
$0.000
Time
300.0s
Tokens
0 tok
#277 Qwen3.8 27B
noneCandidate responses were generated on local hardware. OpenRouter was used only for judging.
Cost
~$0.001Estimated GPU electricity cost only. It excludes the rest of the computer and hardware ownership. NVIDIA GeForce RTX 3090: 0.0066 h × 0.300 kW × €0.2896/kWh × 1.138 USD/EUR = ~$0.000657.
6.9Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
5.8Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
75.0%Attempt pass rate = passed attempts / total attempts across runs.…
2Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.API error: 1Wrong answer: 1Response Time (avg)2.49sResponse Time (max)3.97sResponse Time (total)9.97sA test is fully passed only if every run passed for that test.…
2.49sResponse Time (avg)…
615Total Input Tokens…
204Output Tokens…
1,290Reasoning Tokens…
Qwen3.8 27BCandidate responses were generated on local hardware. OpenRouter was used only for judging.
6.5Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
50.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 2Response Time (avg)828msResponse Time (max)1.95sResponse Time (total)3.31sA test is fully passed only if every run passed for that test.…
4.3Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
7.3Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
11.1%Attempt pass rate = passed attempts / total attempts across runs.…
1Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.API error: 1Extra formatting: 1Wrong answer: 1Response Time (avg)39.07sResponse Time (max)104.98sResponse Time (total)117.21sA test is fully passed only if every run passed for that test.…
39.07sResponse Time (avg)…
6,948Total Input Tokens…
1,166Output Tokens…
55,883Reasoning Tokens…
Qwen3.8 27BCandidate responses were generated on local hardware. OpenRouter was used only for judging.
5.5Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
33.3%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 2Response Time (avg)1.19sResponse Time (max)1.97sResponse Time (total)3.56sA test is fully passed only if every run passed for that test.…
6.5Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
50.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.API error: 1Response Time (avg)74.48sResponse Time (max)118.35sResponse Time (total)148.95sA test is fully passed only if every run passed for that test.…
74.48sResponse Time (avg)…
23,854Total Input Tokens…
485Output Tokens…
28,559Reasoning Tokens…
Qwen3.8 27BCandidate responses were generated on local hardware. OpenRouter was used only for judging.
3.0Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
0.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Invalid tool call: 1Wrong answer: 1Response Time (avg)24.10sResponse Time (max)44.76sResponse Time (total)48.21sA test is fully passed only if every run passed for that test.…
6.5Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
50.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 1Response Time (avg)3.59sResponse Time (max)4.92sResponse Time (total)7.18sA test is fully passed only if every run passed for that test.…
3.59sResponse Time (avg)…
7,776Total Input Tokens…
284Output Tokens…
2,140Reasoning Tokens…
Qwen3.8 27BCandidate responses were generated on local hardware. OpenRouter was used only for judging.
6.5Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
50.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 1Response Time (avg)1.14sResponse Time (max)1.18sResponse Time (total)2.27sA test is fully passed only if every run passed for that test.…
3.0Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
0.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Extra formatting: 2Wrong answer: 1Response Time (avg)7.28sResponse Time (max)8.71sResponse Time (total)21.85sA test is fully passed only if every run passed for that test.…
7.28sResponse Time (avg)…
724Total Input Tokens…
1,101Output Tokens…
7,738Reasoning Tokens…
Qwen3.8 27BCandidate responses were generated on local hardware. OpenRouter was used only for judging.
7.7Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
66.7%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 1Response Time (avg)552msResponse Time (max)675msResponse Time (total)1.66sA test is fully passed only if every run passed for that test.…
5.1Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
0.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 1Response Time (avg)7.10sResponse Time (max)7.10sResponse Time (total)7.10sA test is fully passed only if every run passed for that test.…
7.10sResponse Time (avg)…
516Total Input Tokens…
179Output Tokens…
539Reasoning Tokens…
Qwen3.8 27BCandidate responses were generated on local hardware. OpenRouter was used only for judging.
4.2Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
9.9Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
0.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 1Response Time (avg)1.31sResponse Time (max)1.31sResponse Time (total)1.31sA test is fully passed only if every run passed for that test.…
9.8Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
100.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.No failed answers.Response Time (avg)1.51sResponse Time (max)1.91sResponse Time (total)3.01sA test is fully passed only if every run passed for that test.…
1.51sResponse Time (avg)…
699Total Input Tokens…
90Output Tokens…
675Reasoning Tokens…
Qwen3.8 27BCandidate responses were generated on local hardware. OpenRouter was used only for judging.
6.3Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
50.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 1Response Time (avg)582msResponse Time (max)633msResponse Time (total)1.16sA test is fully passed only if every run passed for that test.…
3.5Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
4.4Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
33.3%Attempt pass rate = passed attempts / total attempts across runs.…
2Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 3Response Time (avg)2.47sResponse Time (max)2.86sResponse Time (total)7.40sA test is fully passed only if every run passed for that test.…
2.47sResponse Time (avg)…
696Total Input Tokens…
285Output Tokens…
1,576Reasoning Tokens…
Qwen3.8 27BCandidate responses were generated on local hardware. OpenRouter was used only for judging.
6.0Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
33.3%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Did not follow instructions: 1Wrong answer: 1Response Time (avg)1.25sResponse Time (max)1.84sResponse Time (total)3.76sA test is fully passed only if every run passed for that test.…
10.0Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
100.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.No failed answers.Response Time (avg)4.77sResponse Time (max)4.77sResponse Time (total)4.77sA test is fully passed only if every run passed for that test.…
4.77sResponse Time (avg)…
8,391Total Input Tokens…
330Output Tokens…
1,082Reasoning Tokens…
Qwen3.8 27BCandidate responses were generated on local hardware. OpenRouter was used only for judging.
10.0Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
100.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.No failed answers.Response Time (avg)2.75sResponse Time (max)2.75sResponse Time (total)2.75sA test is fully passed only if every run passed for that test.…
3.0Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
0.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 1Response Time (avg)21.72sResponse Time (max)21.72sResponse Time (total)21.72sA test is fully passed only if every run passed for that test.…
21.72sResponse Time (avg)…
204Total Input Tokens…
20Output Tokens…
11,028Reasoning Tokens…
Qwen3.8 27BCandidate responses were generated on local hardware. OpenRouter was used only for judging.
3.0Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance.…
10.0Consistency score reflects run-to-run stability (10 = very consistent, even if consistently wrong).…
0.0%Attempt pass rate = passed attempts / total attempts across runs.…
0Flaky tests had mixed outcomes across runs (at least one pass and one fail).…
A test is fully passed only if every run passed for that test.Wrong answer: 1Response Time (avg)703msResponse Time (max)703msResponse Time (total)703msA test is fully passed only if every run passed for that test.…