Codex GPT-5.5
replicate 1
passUDT-only
Answer
6748Workflow
Galaxy UDTPhyKITSciPyBiopythonMAFFTMann-Whitney U
- Total tokens
- 1.34M
- Output
- 8.3K
New token-optimization experiment
Three additional GPT-5.6 Sol Galaxy replicates are available separately. Previous results remain unchanged.
Question
What is the Mann-Whitney U statistic when comparing raw parsimony informative site counts between animals and fungi?
Comparison
Pass counts and token totals across the same three replicates.
| Agent | Galaxy-API code | Open-ended code |
|---|---|---|
| Codex GPT-5.5 | 3/3 replicate 1 replicate 2 replicate 3 4.15M total | 3/3 replicate 1 replicate 2 replicate 3 1.30M total |
| Codex GPT-5.6 Sol | 3/3 replicate 1 replicate 2 replicate 3 22.13M total | 3/3 replicate 1 replicate 2 replicate 3 1.35M total |
| Codex GPT-5.6 Luna | 3/3 replicate 1 replicate 2 replicate 3 41.55M total | 2/3 replicate 1 replicate 2 replicate 3 3.39M total |
| DeepSeek V4 Pro via Codex | 3/3 replicate 1 replicate 2 replicate 3 37.23M total | 3/3 replicate 1 replicate 2 replicate 3 6.57M total |
| DeepSeek V4 Pro via Claude Codesuperseded | 3/3 replicate 1 replicate 2 replicate 3 7.19M total | 1/3 replicate 1 replicate 2 replicate 3 2.96M total |
Execution condition
Galaxy-API code with skills
Codex GPT-5.5
3 primary replicatesCodex GPT-5.5
replicate 2
passUDT-only
Answer
6748Workflow
Galaxy UDTPhyKITMAFFTSciPyBiopythonMann-Whitney U
- Total tokens
- 935.2K
- Output
- 11.3K
Codex GPT-5.5
replicate 3
passUDT + standard tools
Answer
6748Workflow
Galaxy UDTNonparametric Rank TestsPhyKITSciPyBiopythonMAFFT
- Total tokens
- 1.87M
- Output
- 11.7K
Codex GPT-5.6 Sol
3 primary replicatesCodex GPT-5.6 Sol
replicate 1
passUDT + standard tools
Answer
6748Workflow
UnzipCollection Element IdentifiersGalaxy UDTNonparametric Rank TestsMAFFTPhyKIT
- Total tokens
- 6.36M
- Output
- 21.0K
Codex GPT-5.6 Sol
replicate 2
passUDT + standard tools
Answer
6748Workflow
Galaxy UDTNonparametric Rank TestsMAFFTPhyKIT
- Total tokens
- 2.18M
- Output
- 11.0K
Codex GPT-5.6 Sol
replicate 3
pass
Answer
6748Workflow
UnzipPhykit Alignment BasedCollapse DatasetNonparametric Rank TestsMAFFT
- Total tokens
- 13.58M
- Output
- 28.5K
Codex GPT-5.6 Luna
3 primary replicatesCodex GPT-5.6 Luna
replicate 1
passUDT + standard tools
Answer
6748Workflow
UnzipCollection Element IdentifiersPhykit Alignment BasedGalaxy UDTNonparametric Rank TestsMAFFT
- Total tokens
- 26.07M
- Output
- 47.1K
Codex GPT-5.6 Luna
replicate 2
passUDT + standard tools
Answer
6748Workflow
UnzipPhykit Alignment BasedGalaxy UDTNonparametric Rank TestsMAFFT
- Total tokens
- 6.99M
- Output
- 24.5K
Codex GPT-5.6 Luna
replicate 3
passUDT + standard tools
Answer
6748Workflow
UnzipPhykit Alignment BasedGalaxy UDTNonparametric Rank TestsBUSCOMAFFT
- Total tokens
- 8.49M
- Output
- 28.7K
DeepSeek V4 Pro via Codex
3 primary replicatesDeepSeek V4 Pro via Codex
replicate 1
passUDT + standard tools
Answer
6748.0Workflow
UnzipCollection Element IdentifiersFilter From Filemore stepsSciPyMann-Whitney U
- Total tokens
- 27.80M
- Output
- 74.7K
DeepSeek V4 Pro via Codex
replicate 2
passUDT + standard tools
Answer
6748Workflow
Galaxy UDTNonparametric Rank TestsPhyKITMAFFT
- Total tokens
- 2.99M
- Output
- 22.2K
DeepSeek V4 Pro via Codex
replicate 3
passUDT + standard tools
Answer
6748Workflow
Galaxy UDTNonparametric Rank TestsAddvaluemore stepsMAFFTPhyKIT
- Total tokens
- 6.44M
- Output
- 33.0K
DeepSeek V4 Pro via Claude CodeSuperseded
3 primary replicatesDeepSeek V4 Pro via Claude Code
replicate 1
passUDT + standard tools
Answer
6748Workflow
Galaxy UDTNonparametric Rank TestsMAFFTSciPyMann-Whitney U
- Total tokens
- 2.51M
- Output
- 31.2K
DeepSeek V4 Pro via Claude Code
replicate 2
passUDT + standard tools
Answer
6748Workflow
Nonparametric Rank TestsPhyKITMAFFT
- Total tokens
- 2.47M
- Output
- 25.2K
DeepSeek V4 Pro via Claude Code
replicate 3
passUDT + standard tools
Answer
6748Workflow
Galaxy UDTNonparametric Rank TestsPhyKITBiopythonMAFFT
- Total tokens
- 2.20M
- Output
- 24.0K
Execution condition
Open-ended code with skills
Codex GPT-5.5
3 primary replicatesCodex GPT-5.5
replicate 1
pass
Answer
6748.0Workflow
Local analysisMAFFTMann-Whitney USciPyBiopythonBUSCO
- Total tokens
- 382.3K
- Output
- 8.6K
Codex GPT-5.5
replicate 2
pass
Answer
6748Workflow
Local analysisBUSCOMAFFTSciPyMann-Whitney Umean summary
- Total tokens
- 723.7K
- Output
- 9.2K
Codex GPT-5.5
replicate 3
pass
Answer
6748Workflow
Local analysisMAFFTSciPyMann-Whitney U
- Total tokens
- 190.7K
- Output
- 6.2K
Codex GPT-5.6 Sol
3 primary replicatesCodex GPT-5.6 Sol
replicate 1
pass
Answer
6748Workflow
Local analysis
- Total tokens
- 245.4K
- Output
- 7.0K
Codex GPT-5.6 Sol
replicate 2
pass
Answer
6748Workflow
Local analysis
- Total tokens
- 730.9K
- Output
- 6.7K
Codex GPT-5.6 Sol
replicate 3
pass
Answer
6748Workflow
Local analysis
- Total tokens
- 375.9K
- Output
- 7.0K
Codex GPT-5.6 Luna
3 primary replicatesCodex GPT-5.6 Luna
replicate 1
fail
Why it missed
Gap characters were counted as statesThis run treated alignment gaps as character states when counting informative sites and obtained U=6397. The PhyKIT-compatible calculation treats gaps as missing and gives U=6748 on the same 241 and 255 alignments.
Answer
6397Workflow
Local analysis
- Total tokens
- 640.1K
- Output
- 16.3K
Codex GPT-5.6 Luna
replicate 2
pass
Answer
6748Workflow
Local analysis
- Total tokens
- 2.03M
- Output
- 19.6K
Codex GPT-5.6 Luna
replicate 3
pass
Answer
6748Workflow
Local analysis
- Total tokens
- 722.9K
- Output
- 15.6K
DeepSeek V4 Pro via Codex
3 primary replicatesDeepSeek V4 Pro via Codex
replicate 1
pass
Answer
6748.0Workflow
Local analysisBUSCOMAFFTPhyKITMann-Whitney USciPy
- Total tokens
- 3.89M
- Output
- 30.3K
DeepSeek V4 Pro via Codex
replicate 2
pass
Answer
6748.0Workflow
Local analysisBUSCOMAFFTSciPyBiopythonMann-Whitney U
- Total tokens
- 1.95M
- Output
- 21.8K
DeepSeek V4 Pro via Codex
replicate 3
pass
Answer
6748.0Workflow
Local analysisMAFFTSciPyMann-Whitney U
- Total tokens
- 731.5K
- Output
- 9.6K
DeepSeek V4 Pro via Claude CodeSuperseded
3 primary replicatesDeepSeek V4 Pro via Claude Code
replicate 1
pass
Answer
6748.0Workflow
Local analysisBUSCOMAFFTSciPy
- Total tokens
- 823.2K
- Output
- 16.7K
DeepSeek V4 Pro via Claude Code
replicate 2
fail
Why it missed
Statistic orientation errorThis run returned the opposite Mann-Whitney orientation, 55058, and counted gaps as characters. Using animals as the first group gives U=6748.
Answer
55058Workflow
Local analysisBUSCOMAFFTSciPy
- Total tokens
- 1.26M
- Output
- 31.6K
DeepSeek V4 Pro via Claude Code
replicate 3
fail
Why it missed
Data scope changedThis run reduced both groups to 241 shared genes and obtained U=6181. The requested full 241-animal versus 255-fungal comparison gives U=6748.
Answer
6181.0Workflow
Local analysisMAFFTSciPyMann-Whitney U
- Total tokens
- 872.5K
- Output
- 13.5K
bix-12-q6 · October 2026 · three additional replicates
GPT-5.6 Sol · token optimization
Galaxy-API code, high reasoning, Fast service. Previous runs are retained above.
Galaxy-API code · October 2026
replicate 1
Correct
Answer
6748.0- Total tokens
- 1.60M
- Input
- 1.58M
- Cached input
- 1.52M
- Output
- 13.4K
Galaxy-API code · October 2026
replicate 2
Correct
Answer
6748- Total tokens
- 897.4K
- Input
- 888.2K
- Cached input
- 846.3K
- Output
- 9.2K
Galaxy-API code · October 2026
replicate 3
Correct
Answer
6748- Total tokens
- 2.07M
- Input
- 2.06M
- Cached input
- 1.99M
- Output
- 9.6K
Open-ended code averaged 0.45M tokens; previous website Galaxy averaged 7.38M; the new Galaxy runs averaged 1.52M, or 3.37× Open-ended code. Each mean covers three replicates of this same task with GPT-5.6 Sol.
Cached input is already included in input tokens. Full nested tool/job evidence is in each HF run's galaxy_execution_records.tar.gz, with an accompanying member index.