Meta Muse Spark 1.1 Report - ProtocolQA: leaderboard
LAB-Bench ProtocolQA: debugging wet-lab protocols modified to introduce errors. Meta Muse Spark 1.1 Evaluation Report (July 2026), Table 1; accuracy in percent, higher is more capable. Muse Spark 1.1 is the helpful-only variant, per Section 2.1.1; Claude Opus 4.8 is not reported because of refusals. The report's Muse Spark 1.0 cell used an extra abstention choice and is not included.
Metric: Accuracy (%). Source: research.meta.ai. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 88.9 | #64 (Gemini 3.1 Pro (Preview)) |
| 2 | GPT-5.5 (xHigh) | 78.2 | #33 (GPT-5.5) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=meta-muse-spark-1-1-report-protocolqa · How It Works · Data refreshed daily, snapshot 2026-11-05.