bix-6-q4 | Current BixBench results

New token-optimization experiment

Three additional GPT-5.6 Sol Galaxy replicates are available separately. Previous results remain unchanged.

View the October 2026 batch and its run records

Question

What is the Spearman correlation coefficient between the replicate MAGeCK P-values for chronic round 1?

Accepted answer: (0.001,0.09)
Domain: general

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 5.66M total	3/3 replicate 1 replicate 2 replicate 3 503.6K total
Codex GPT-5.6 Sol	3/3 replicate 1 replicate 2 replicate 3 6.44M total	3/3 replicate 1 replicate 2 replicate 3 501.7K total
Codex GPT-5.6 Luna	3/3 replicate 1 replicate 2 replicate 3 18.74M total	3/3 replicate 1 replicate 2 replicate 3 399.1K total
DeepSeek V4 Pro via Codex	3/3 replicate 1 replicate 2 replicate 3 8.87M total	3/3 replicate 1 replicate 2 replicate 3 476.4K total
DeepSeek V4 Pro via Claude Codesuperseded	3/3 replicate 1 replicate 2 replicate 3 5.31M total	3/3 replicate 1 replicate 2 replicate 3 760.8K total

Execution condition

Galaxy-API code with skills

Codex GPT-5.5

3 primary replicates

Codex GPT-5.5

replicate 1

passUDT + standard tools

Answer0.02097098108581123

Workflow

Excel to tabularSelect columnsSpearman correlationRename columnGalaxy UDTSciPy

Total tokens: 2.60M
Output: 14.8K

Galaxy history Run trace

Codex GPT-5.5

replicate 2

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsSpearman correlationReplace column values

Total tokens: 1.73M
Output: 7.4K

Galaxy history Run trace

Codex GPT-5.5

replicate 3

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsRemove headerSpearman correlation

Total tokens: 1.33M
Output: 6.4K

Galaxy history Run trace

Codex GPT-5.6 Sol

3 primary replicates

Codex GPT-5.6 Sol

replicate 1

pass

Answer0.020970981085811226

Workflow

Excel to tabularTp Cut ToolSelect columnsSpearman correlationRegex1

Total tokens: 1.69M
Output: 10.5K

Galaxy history Run trace

Codex GPT-5.6 Sol

replicate 2

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsReplace textSpearman correlationRemove headerAdd Line To File

Total tokens: 2.21M
Output: 8.4K

Galaxy history Run trace

Codex GPT-5.6 Sol

replicate 3

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsSpearman correlationTp Find And ReplaceSelect trailing rowsAdd Line To File

Total tokens: 2.54M
Output: 9.4K

Galaxy history Run trace

Codex GPT-5.6 Luna

3 primary replicates

Codex GPT-5.6 Luna

replicate 1

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsSpearman correlationRegexcolumn1Rename column

Total tokens: 9.05M
Output: 29.1K

Galaxy history Run trace

Codex GPT-5.6 Luna

replicate 2

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsRename columnFilter TabularSpearman correlationTp Grep Tool

Total tokens: 5.98M
Output: 22.8K

Galaxy history Run trace

Codex GPT-5.6 Luna

replicate 3

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsSpearman correlationRename columnReplace text

Total tokens: 3.71M
Output: 23.3K

Galaxy history Run trace

DeepSeek V4 Pro via Codex

3 primary replicates

DeepSeek V4 Pro via Codex

replicate 1

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsSpearman correlationReplace textRemove header

Total tokens: 3.83M
Output: 24.1K

Galaxy history Run trace

DeepSeek V4 Pro via Codex

replicate 2

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsSpearman correlationTp Find And Replace

Total tokens: 2.39M
Output: 14.8K

Galaxy history Run trace

DeepSeek V4 Pro via Codex

replicate 3

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsSpearman correlationReplace text

Total tokens: 2.65M
Output: 16.7K

Galaxy history Run trace

DeepSeek V4 Pro via Claude CodeSuperseded

3 primary replicates

DeepSeek V4 Pro via Claude Code

replicate 1

passUDT + standard tools

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsSpearman correlationGalaxy UDTSciPy

Total tokens: 2.52M
Output: 22.4K

Galaxy history Run trace

DeepSeek V4 Pro via Claude Code

replicate 2

pass

Answer0.020971

Workflow

Spearman correlation

Total tokens: 1.36M
Output: 12.5K

Galaxy history Run trace

DeepSeek V4 Pro via Claude Code

replicate 3

pass

Answer0.020970981085811226

Workflow

Excel to tabularSelect columnsSpearman correlationReplace textSelect trailing rows

Total tokens: 1.43M
Output: 14.7K

Galaxy history Run trace

Execution condition

Open-ended code with skills

Codex GPT-5.5

3 primary replicates

Codex GPT-5.5

replicate 1

pass

Answer0.02097098108581123

Workflow

Local analysisSciPySpearman correlation

Total tokens: 178.0K
Output: 3.1K

Run trace

Codex GPT-5.5

replicate 2

pass

Answer0.020970981085811229

Workflow

Local analysisSciPySpearman correlation

Total tokens: 153.4K
Output: 3.3K

Run trace

Codex GPT-5.5

replicate 3

pass

Answer0.020970981085811229

Workflow

Local analysisSciPySpearman correlation

Total tokens: 172.2K
Output: 3.4K

Run trace

Codex GPT-5.6 Sol

3 primary replicates

Codex GPT-5.6 Sol

replicate 1

pass

Answer0.02097098108581123

Workflow

Local analysis

Total tokens: 152.7K
Output: 2.9K

Run trace

Codex GPT-5.6 Sol

replicate 2

pass

Answer0.02097098108581123

Workflow

Local analysis

Total tokens: 158.9K
Output: 2.8K

Run trace

Codex GPT-5.6 Sol

replicate 3

pass

Answer0.02097098108581123

Workflow

Local analysis

Total tokens: 190.1K
Output: 3.6K

Run trace

Codex GPT-5.6 Luna

3 primary replicates

Codex GPT-5.6 Luna

replicate 1

pass

Answer0.0209709810858112

Workflow

Local analysis

Total tokens: 145.6K
Output: 3.8K

Run trace

Codex GPT-5.6 Luna

replicate 2

pass

Answer0.0209709810858112

Workflow

Local analysis

Total tokens: 131.7K
Output: 3.5K

Run trace

Codex GPT-5.6 Luna

replicate 3

pass

Answer0.036586792816685108

Workflow

Local analysis

Total tokens: 121.8K
Output: 2.9K

Run trace

DeepSeek V4 Pro via Codex

3 primary replicates

DeepSeek V4 Pro via Codex

replicate 1

pass

Answer0.020970981085811226

Workflow

Local analysisSciPySpearman correlation

Total tokens: 178.2K
Output: 4.3K

Run trace

DeepSeek V4 Pro via Codex

replicate 2

pass

Answer0.020970981085811

Workflow

Local analysisSciPySpearman correlation

Total tokens: 116.8K
Output: 2.9K

Run trace

DeepSeek V4 Pro via Codex

replicate 3

pass

Answer0.02097098108581123

Workflow

Local analysisSciPySpearman correlationstatsmodels

Total tokens: 181.3K
Output: 4.5K

Run trace

DeepSeek V4 Pro via Claude CodeSuperseded

3 primary replicates

DeepSeek V4 Pro via Claude Code

replicate 1

pass

Answer0.020971

Workflow

Local analysisSciPySpearman correlation

Total tokens: 199.6K
Output: 2.3K

Run trace

DeepSeek V4 Pro via Claude Code

replicate 2

pass

Answer0.020970981085811

Workflow

Local analysisSciPySpearman correlation

Total tokens: 298.2K
Output: 2.2K

Run trace

DeepSeek V4 Pro via Claude Code

replicate 3

pass

Answer0.020970981085811226

Workflow

Local analysisSciPySpearman correlationmean summary

Total tokens: 263.0K
Output: 2.0K

Run trace

bix-6-q4 · 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

Answer0.020970981085811226

Total tokens: 630.4K
Input: 625.6K
Cached input: 593.0K
Output: 4.8K

Run trace Final answer Galaxy history

Galaxy-API code · October 2026

replicate 2

Correct

Answer0.020970981085811226

Total tokens: 923.4K
Input: 916.2K
Cached input: 879.1K
Output: 7.2K

Run trace Final answer Galaxy history

Galaxy-API code · October 2026

replicate 3

Correct

Answer0.020970981085811226

Total tokens: 800.1K
Input: 793.8K
Cached input: 757.4K
Output: 6.4K

Run trace Final answer Galaxy history

Open-ended code averaged 0.17M tokens; previous website Galaxy averaged 2.15M; the new Galaxy runs averaged 0.78M, or 4.69× 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.

Optimization overview and all 150 runs