Success rate (protocol-development pass@5: probability that at least one of 5 attempts passes; best-of-k, never comparable with single-attempt pass rate; top reasoning tier, the highest of this system's three settings) of Gemini 3.8 Flash in Gemini CLI 0.63.0, high reasoning effort on BenchBench-Cloning protocol-development tasks (50)
10.5% Success rate (protocol-development pass@5: probability that at least one of 5 attempts passes; best-of-k, never comparable with single-attempt pass rate; top reasoning tier, the highest of this system's three settings)
Methods
- Tested configuration
- Gemini 3.8 Flash in Gemini CLI 0.63.0, high reasoning effort
- Benchmark
- BenchBench-Cloning protocol-development set
- Dataset subset
- BenchBench-Cloning protocol-development tasks (50)
- Evaluation
- Gemini 3.8 Flash on BenchBench-Cloning protocol development, pass@5
- Coverage
- Not reported scored / Not reported eligible
- Uncertainty
- 95% CI 4.0 to 18.1. 95% confidence intervals from 20,000 bootstrap resamples of the tasks, retaining all attempts of each sampled task (Section 3.4).
- Evidence
- Author-reported evaluation · source checkedBenchBench-Cloning: Evaluating Molecular Cloning Design and Protocol Development (arXiv:2610.33355v1) · Table G1, row 'Gemini 3.8 Flash', column 'pass@5 % [95% CI]'
A source-checked result verifies the numerical transcription, not every model or protocol detail. Evaluation metadata: source checked. Source checked does not mean independently reproduced.
Reproduction
- Split
- Evaluation only; all 50 tasks scored
- Adaptation
- Single-turn agent attempt: the agent inspects the supplied files, uses the available tools, revises within the attempt and writes its submission to a designated answer file; no clarifying questions and no feedback between attempts
- Scoring implementation
- pass@5: probability that at least one of 5 attempts passes the simulator
No execution recipe has been verified for this exact configuration and evaluation. A benchmark's general instructions may use different inputs, splits or model settings.
Reproducing this published result requires matching its model configuration, data, split and scorer. Source checking or a successful smoke test does not establish score reproduction.
Evidence
Source checking verifies the cited claim or transcription. It does not establish independent reproduction.
Evidence table
Inspect claims, sources and review details
Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.
One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.
1 evidence row matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
| Reported result 10.5 Individual claims | BenchBench-Cloning: Evaluating Molecular Cloning Design and Protocol Development (arXiv:2610.33355v1) Table G1, row 'Gemini 3.8 Flash', column 'pass@5 % [95% CI]' Version: arXiv v1, dated 23 October 2026 | source checked ["independent-cell-check","deterministic-table-parse","pdf-text-parse","source-hash-verification"] · 2026-11-06T00:00:00Z author reported Source has a recorded evidence concern. Consult its source page before using the claim. Audit detailsField: Source artifact SHA-256: Hash scope: Hash scope not separately documented; inspect source record Archive member: artifacts/arxiv-2610.33355v1.layout.txt.gz Extraction artifact SHA-256: |
Sources and history
Release 2026-11-21-c8bea25e5abb · Record review: source checked
1 source record and release history
- BenchBench-Cloning: Evaluating Molecular Cloning Design and Protocol Development (arXiv:2610.33355v1) · Original source · arXiv v1, dated 23 October 2026
Technical metadata and extraction receipts
Stable ID: benchbench-cloning-20261106-result-gemini-3-8-flash-pass-at-5
- denominator
- 500
- metric
- success-rate
- metric direction
- higher
- metric qualifier
- protocol-development pass@5: probability that at least one of 5 attempts passes; best-of-k, never comparable with single-attempt pass rate; top reasoning tier, the highest of this system's three settings
- numeric value
- 10.5
- printed value
- 10.5
- review
- artifact sha256: 6f5044b1bbbd81ad86b23cb536e2cae6428e0d9126f11546041355518febdc39; method: independent-cell-check; deterministic-table-parse; pdf-text-parse; source-hash-verification; method note: Independent review pass. All three pinned artifacts were re-downloaded and their SHA-256 checked against the source records; the archived gzipped text layer matches its recorded hash and decompresses byte for byte to a pdftotext -layout run on the refetched PDF. The repository 404 was rechecked at review time on both the web page and the GitHub REST API. npm run batch -- check-values reproduced the extractor's figures. All 224 results were then re-read by a reviewer script written for this pass, which locates each table body by its own printed header line rather than by caption proximity, takes the column order from that header and splits each row on runs of two or more spaces, without using extract/parse_tables.py or extract/check_tables.py; every printed value and every printed interval matched its own field. Three arithmetic constraints were re-derived independently: each Table G2 column sums to the printed top-effort construct-design numerator over 75, each Table G1 Solved-at-least-once count over 50 equals the printed pass@10, and every Table D2 aggregate equals (50 x protocol + 15 x construct) / 65. The 1/75 grid check that identifies the single off-grid rate was recomputed over all 24 lanes using each printed value's own rounding window.; retrieval url: https://arxiv.org/pdf/2610.33355v1; reviewed at: 2026-11-06T00:00:00Z; reviewer: claude; reviewer note: Claude research agent; no human review claimed. Reviewed by a different agent from the one that extracted the batch. Transcription checked against the pinned source; nothing was reproduced and no value was recomputed.
- source locator
- Table G1, row 'Gemini 3.8 Flash', column 'pass@5 % [95% CI]'
- source warnings
- Table G1 records a k = 9 cap (4 tasks) for this system: some tasks have fewer than ten attempts, and the paper does not say how pass@k handles the shortfall.
- uncertainty
- level: 0.95; lower: 4.0; method: bootstrap; note: 95% confidence intervals from 20,000 bootstrap resamples of the tasks, retaining all attempts of each sampled task (Section 3.4).; printed: 10.5 [4.0, 18.1]; resamples: 20000; type: confidence_interval; unit: percent; upper: 18.1
- unit
- percent