PHREEQC-MCQ-200: leaderboard
PHREEQC-MCQ-200 (University of California, Riverside and Irvine, 2026) asks tool-using LLM agents 200 multiple-choice questions whose answers come from running the PHREEQC aqueous-geochemistry simulator: the agent writes the simulator input, runs it and reads the output within 24 steps. Accuracy in percent over all 200 items, higher is better; this board uses the table-of-contents output interface, in which the agent fetches output sections on demand.
Metric: Accuracy (%; 200 four-option questions from 21 validated PHREEQC aqueous-geochemistry simulation scenarios; the custom agent writes a PHREEQC input, runs the simulator and reads its output within 24 steps, then writes one answer letter; a missing or unparseable answer counts as wrong; table-of-contents output access, Agent-TOC). Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 84 | #161 |
| 2 | Claude Opus 4.6 | 83.5 | #102 |
| 3 | GPT-5.4 | 72.5 | #123 |
| 4 | GPT-5.2 | 54 | #176 |
| 5 | GPT-5.1 | 54 | #229 |
| 6 | Gemini 3 Flash (Preview) | 23.5 | #144 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=phreeqc-mcq-200 · How It Works · Data refreshed daily, snapshot 2026-11-13.