bix-28-q3 | 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

Using PhyKIT to calculate long branch scores, what is the median long branch score for the fungal gene 996662at2759?

Accepted answer: -30.4551
Domain: phylogenetics

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 5.03M total	3/3 replicate 1 replicate 2 replicate 3 1.64M total
Codex GPT-5.6 Sol	2/3 replicate 1 replicate 2 replicate 3 4.02M total	3/3 replicate 1 replicate 2 replicate 3 1.56M total
Codex GPT-5.6 Luna	3/3 replicate 1 replicate 2 replicate 3 8.06M total	3/3 replicate 1 replicate 2 replicate 3 2.31M total
DeepSeek V4 Pro via Codex	3/3 replicate 1 replicate 2 replicate 3 10.93M total	3/3 replicate 1 replicate 2 replicate 3 2.80M total
DeepSeek V4 Pro via Claude Codesuperseded	3/3 replicate 1 replicate 2 replicate 3 3.91M total	3/3 replicate 1 replicate 2 replicate 3 2.96M total

Execution condition

Galaxy-API code with skills

Codex GPT-5.5

3 primary replicates

Codex GPT-5.5

replicate 1

passUDT + standard tools

Answer-30.4551

Workflow

Phykit MetricsGalaxy UDTMAFFTNewick analysismedian summary

Total tokens: 1.70M
Output: 13.0K

Galaxy history Run trace

Codex GPT-5.5

replicate 2

pass

Answer-30.45505

Workflow

Phykit MetricsMAFFTNewick analysis

Total tokens: 1.40M
Output: 7.1K

Galaxy history Run trace

Codex GPT-5.5

replicate 3

passUDT + standard tools

Answer-30.4551

Workflow

Phykit MetricsGalaxy UDTMAFFTmedian summary

Total tokens: 1.93M
Output: 11.0K

Galaxy history Run trace

Codex GPT-5.6 Sol

3 primary replicates

Codex GPT-5.6 Sol

replicate 1

fail

Why it missed

Wrong PhyKIT output was treated as the medianGalaxy produced the four requested per-taxon long-branch scores, whose median is -30.4551. The agent then switched the wrapper to non-verbose mode and accepted its unrelated scalar 146.3023 instead of reducing the verified per-taxon rows.

Answer146.3023

Workflow

Phykit MetricsBUSCOMAFFT

Total tokens: 1.81M
Output: 9.8K

Galaxy history Run trace

Codex GPT-5.6 Sol

replicate 2

pass

Answer-30.45505

Workflow

Phykit Metrics

Total tokens: 1.09M
Output: 6.1K

Galaxy history Run trace

Codex GPT-5.6 Sol

replicate 3

pass

Answer-30.45505

Workflow

Phykit Metricsmedian summary

Total tokens: 1.12M
Output: 7.1K

Galaxy history Run trace

Codex GPT-5.6 Luna

3 primary replicates

Codex GPT-5.6 Luna

replicate 1

pass

Answer-30.45505

Workflow

Phykit MetricsMAFFTIQ-TREE

Total tokens: 2.51M
Output: 18.9K

Galaxy history Run trace

Codex GPT-5.6 Luna

replicate 2

pass

Answer-30.45505

Workflow

Phykit MetricsMAFFTNewick analysis

Total tokens: 2.66M
Output: 15.6K

Galaxy history Run trace

Codex GPT-5.6 Luna

replicate 3

pass

Answer-30.45505

Workflow

Phykit MetricsMAFFTmedian summary

Total tokens: 2.90M
Output: 19.0K

Galaxy history Run trace

DeepSeek V4 Pro via Codex

3 primary replicates

DeepSeek V4 Pro via Codex

replicate 1

pass

Answer-30.45505

Workflow

Phykit MetricsMAFFT

Total tokens: 2.41M
Output: 23.2K

Galaxy history Run trace

DeepSeek V4 Pro via Codex

replicate 2

pass

Answer-30.4551

Workflow

Phykit MetricsMAFFTNewick analysis

Total tokens: 6.17M
Output: 35.7K

Galaxy history Run trace

DeepSeek V4 Pro via Codex

replicate 3

pass

Answer-30.45505

Workflow

Phykit MetricsMAFFTNewick analysis

Total tokens: 2.35M
Output: 19.5K

Galaxy history Run trace

DeepSeek V4 Pro via Claude CodeSuperseded

3 primary replicates

DeepSeek V4 Pro via Claude Code

replicate 1

pass

Answer-30.45505

Workflow

Phykit MetricsMAFFTmedian summary

Total tokens: 1.49M
Output: 9.3K

Galaxy history Run trace

DeepSeek V4 Pro via Claude Code

replicate 2

pass

Answer-30.45505

Workflow

Phykit MetricsMAFFTNewick analysis

Total tokens: 1.80M
Output: 19.1K

Galaxy history Run trace

DeepSeek V4 Pro via Claude Code

replicate 3

pass

Answer-30.45505

Workflow

Phykit Metricsmedian summary

Total tokens: 620.7K
Output: 6.2K

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

Answer-30.4551

Workflow

Local analysisNewick analysisMAFFTPhyKITIQ-TREE

Total tokens: 737.4K
Output: 3.6K

Run trace

Codex GPT-5.5

replicate 2

pass

Answer-30.455089045661694

Workflow

Local analysisBUSCOPhyKITMAFFT

Total tokens: 446.8K
Output: 3.6K

Run trace

Codex GPT-5.5

replicate 3

pass

Answer-30.455089045661694

Workflow

Local analysisBUSCOPhyKITMAFFTIQ-TREENewick analysis

Total tokens: 457.9K
Output: 6.0K

Run trace

Codex GPT-5.6 Sol

3 primary replicates

Codex GPT-5.6 Sol

replicate 1

pass

Answer-30.4551

Workflow

Local analysis

Total tokens: 515.2K
Output: 3.9K

Run trace

Codex GPT-5.6 Sol

replicate 2

pass

Answer-30.4551

Workflow

Local analysis

Total tokens: 387.9K
Output: 3.5K

Run trace

Codex GPT-5.6 Sol

replicate 3

pass

Answer-30.4551

Workflow

Local analysis

Total tokens: 656.7K
Output: 3.9K

Run trace

Codex GPT-5.6 Luna

3 primary replicates

Codex GPT-5.6 Luna

replicate 1

pass

Answer-30.4551

Workflow

Local analysis

Total tokens: 567.8K
Output: 6.6K

Run trace

Codex GPT-5.6 Luna

replicate 2

pass

Answer-30.4551

Workflow

Local analysis

Total tokens: 582.5K
Output: 6.8K

Run trace

Codex GPT-5.6 Luna

replicate 3

pass

Answer-30.4551

Workflow

Local analysis

Total tokens: 1.16M
Output: 9.8K

Run trace

DeepSeek V4 Pro via Codex

3 primary replicates

DeepSeek V4 Pro via Codex

replicate 1

pass

Answer-30.4551

Workflow

Local analysisPhyKITBUSCOMAFFTIQ-TREEmean summary

Total tokens: 1.63M
Output: 17.6K

Run trace

DeepSeek V4 Pro via Codex

replicate 2

pass

Answer-30.4551

Workflow

Local analysisBUSCOPhyKITMAFFTIQ-TREE

Total tokens: 574.1K
Output: 7.5K

Run trace

DeepSeek V4 Pro via Codex

replicate 3

pass

Answer-30.4551

Workflow

Local analysisBUSCOPhyKITSciPyMAFFT

Total tokens: 597.9K
Output: 7.2K

Run trace

DeepSeek V4 Pro via Claude CodeSuperseded

3 primary replicates

DeepSeek V4 Pro via Claude Code

replicate 1

pass

Answer-30.4551

Workflow

Local analysisBUSCOPhyKITMAFFTmedian summarymean summary

Total tokens: 547.5K
Output: 8.9K

Run trace

DeepSeek V4 Pro via Claude Code

replicate 2

pass

Answer-30.45505

Workflow

Local analysisBUSCOPhyKITMAFFTNewick analysis

Total tokens: 743.4K
Output: 11.0K

Run trace

DeepSeek V4 Pro via Claude Code

replicate 3

pass

Answer-30.4551

Workflow

Local analysisPhyKITMAFFTmedian summaryIQ-TREENewick analysis

Total tokens: 1.67M
Output: 13.7K

Run trace

bix-28-q3 · 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-30.4551

Total tokens: 638.0K
Input: 632.7K
Cached input: 587.8K
Output: 5.4K

Run trace Final answer Galaxy history

Galaxy-API code · October 2026

replicate 2

Incorrect

Answer146.3023

Total tokens: 619.7K
Input: 614.0K
Cached input: 573.1K
Output: 5.7K

Run trace Final answer Galaxy history

Galaxy-API code · October 2026

replicate 3

Correct

Answer-30.45505

Total tokens: 720.3K
Input: 715.5K
Cached input: 679.7K
Output: 4.8K

Run trace Final answer Galaxy history

Open-ended code averaged 0.52M tokens; previous website Galaxy averaged 1.34M; the new Galaxy runs averaged 0.66M, or 1.27× 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