Molecular cloning is one of the most routine wet-lab workflows, yet existing evaluations of language models in biology mostly test knowledge recall or isolated steps rather than end-to-end experimental planning. We introduce BenchBench-Cloning, a benchmark for large language models comprising 65 molecular cloning tasks grounded in sequence records, laboratory notebooks, and published experimental studies. Planning a molecular cloning experiment requires defining a set of constructs needed for a scientific objective and a construction route to produce them. BenchBench-Cloning contains 50 protocol-development tasks, in which agents produce an executable protocol for a specified construct, and 15 construct-design tasks, in which agents determine the constructs a scientific objective calls for. Protocol-development tasks are evaluated by execution in a cloning simulator; construct-design tasks are evaluated with task-specific deterministic checks and, where needed, rubric-based LLM-as-a-judge evaluation. Across eight systems with repeated attempts per task, GPT-6 Astra scores highest at 30.6%, with the others between 8.6% and 22.2%, and the benchmark remains unsaturated under best-of-ten sampling. As integrating individual cloning decisions into complete, valid plans remains challenging even for frontier models, we expect BenchBench-Cloning to serve as a useful benchmark for tracking progress in experimental planning.
Figures & tables
Figure 1: The two task formats. In protocol development (A) the agent writes a protocol that the simulator executes; in construct design (B) the agent returns constructs and a plan that deterministic checks and, on ten tasks, a model judge score.
Dominant constraint
Tasks
Share
Reasoning the task requires
Methylation state (unknown host / Dam-Dcm sites)
11
22%
Starting plasmid grown in an unknown or mixed host, or a required site overlapping a Dam/Dcm motif; establish a known methylation state or avoid the blocked site before cutting
Homology-directed and seamless assembly
7
14%
Gibson, In-Fusion, in vivo assembly or recombineering with homology arms near the minimum length; in vitro transcription templates needing an oriented promoter
Repeats and tandem arrays
7
14%
Repeated elements (tandem tags, repeated promoters or terminators, homopolymer tracts) where a primer or homology arm binds in more than one place or where a homopolymer run makes length unstable through slippage
Backbone end chemistry and orientation
6
12%
Self-compatible vector ends that re-circularize or dimerize unless dephosphorylated; ends that must be filled or blunted; inserts whose blunt or palindromic ends give both orientations
Golden Gate / MoClo
5
10%
Assemblies with internal Type IIS sites to domesticate and overhang sets that must stay unique
GC-rich templates
4
8%
A window over the disclosed GC limit so PCR through it fails; the fragment must be released or built another way
Table 1: Protocol-development task families. Task counts and shares are shown for each dominant constraint with the reasoning it requires.
Figure 2: Aggregate score by system. Task-weighted means at top effort are shown with 95% confidence intervals.
Figure 3: Pass rate by task set. Means at top effort are shown with 95% confidence intervals for protocol development (A) and construct design (B).
Figure 4: Protocol-development pass@ k . Probability that at least one of k attempts passes is shown with 95% confidence bands for the three leading systems.
Figure 5: Aggregate score against cost per attempt. Each system is shown at its three reasoning settings with the top setting filled; the dashed line is the cost frontier.
Figure 6: Construct-design solves by task and system. Cells give the number of five attempts that passed at top effort.
Figure 7: The dual-luciferase frame-panel task. The request, the five reference constructs, and the number of the 39 submitted panels containing each construct are shown.
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.
Appendix
#
Task
Source
Experimental records supplied
Grading
1
MPRA construct rebuild
Dao et al. 2025, Nat. Commun.
Sequencing reads from the plasmid pool and cellular RNA
Deterministic
2
Matched promoter panel
Picó et al. 2026, Nat. Commun.
Per-well growth and GFP measurements, part map
Deterministic
3
Sodium-channel promoter repair
J. Biol. Chem. 2021, cryptic prokaryotic promoters in neuronal sodium channels
Gel and fragment-stability images, sequence changes, start-site and electrophysiology data
Deterministic
4
Detector choice from antibody data
PLoS ONE 2021, CD55 C81A reporter
Antibody and localisation images, oligonucleotide and genotype tables
Deterministic + rubric judge
5
Reporter choice from data
Ghodke et al. 2016, Biophys. J.
Published microscopy and calibration figure panels
Deterministic
6
Dual-luciferase frame panel
Annibaldis et al. 2020, Nucleic Acids Res.
Published readthrough measurements (three panels)
Deterministic
Appendix
Table A1: Construct-design tasks. Source, experimental records supplied, and grading route are listed for each task.
System
Model id
Harness (version)
Access route
Reasoning settings
GPT-6 Astra
gpt-6-astra
Codex CLI 0.156.1
OpenAI API
low / high / max
GPT-6.1 Sol
gpt-6.1-sol
Codex CLI 0.156.1
OpenAI API
low / high / max
Claude Opus 5.5
claude-opus-5-5
Claude Code 2.1.289
Anthropic API (WIF)
low / high / max
Gemini 4 Argon
gemini-4-argon
Gemini CLI 0.63.0, in-sandbox proxy
Vertex AI (global endpoint)
low / medium / high
Claude Mythos 5.1
claude-mythos-5-1
Claude Code 2.1.289
Anthropic API (WIF)
low / high / max
Grok 4.7
grok-4.7
Grok Build 1.0.46
xAI API
low / high / xhigh
Appendix
Table B1: Model and harness configurations. Model identifier, harness, access route, and reasoning settings are listed per system.
Slice
Trials graded
Agreement
Cohen’s kappa
All trials
387
89.7%
0.76
Tasks with experimental records (6)
137
99.3%
0.98
Construction tasks (9)
250
84.4%
0.65
GPT-6 Astra
62
85.5%
0.72
GPT-6.1 Sol
63
84.1%
0.69
Claude Opus 5.5
74
85.1%
0.67
Appendix
Table C1: Agreement between the Claude Mythos 5.1 judge and reported outcomes. Trials graded, agreement, and Cohen’s kappa are shown per slice.
Figure E1: First blocking failure of protocol-development attempts at top effort. Shares of graded attempts are shown per system, with counts in Table E1 .
Adjudication outcome
Top-tier trials (8 × 75)
All 24 lanes (24 × 75)
Shipped grader result stands
552
1626
A: passes after value-preserving re-expression of the record
17
43
A ′ : engine source checks verify product and route as written
5
7
B: passes after manual verification of the engine evidence
5
26
0: still fails after allowed re-expression
21
95
Unscored (harness failure)
0
3
Appendix
Table F1: Adjudication outcomes for construct-design submissions. Counts are shown for the eight top-effort lanes and for all 24 lanes.
System
Task
Deterministic
Final
Basis
Claude Mythos 5.1
AG/AP1 chimera plasmid
pass
pass
passes after the record is re-expressed without changing any value
Claude Opus 5.5
AG/AP1 chimera plasmid
pass
pass
passes after the record is re-expressed without changing any value
Gemini 4 Argon
AG/AP1 chimera plasmid
pass
pass
passes on manual check of the grader’s evidence
Gemini 4 Argon
AG/AP1 chimera plasmid
pass
pass
passes on manual check of the grader’s evidence
Gemini 4 Argon
AG/AP1 chimera plasmid
fail
fail
still fails after allowed re-expression
Gemini 4 Argon
AG/AP1 chimera plasmid
pass
pass
passes on manual check of the grader’s evidence
Appendix
Table F2: Reviewed construct-design attempts at top effort. Deterministic results after review and final results at the 0.9 threshold are listed with the basis for each decision.