SciConvBench - Disambiguation - PDEs - Final Resolution: leaderboard
SciConvBench (2026) asks an assistant to turn an ill-posed computational-science request into a correct task specification by questioning a simulated scientist, one clarification per turn, judged against the annotated missing or conflicting requirements. This board is the final resolution rate on pdes disambiguation cases (requests with missing information (boundary conditions, material properties, geometry, solver settings or target outputs removed)), the share whose final specification resolves every issue; percent, higher is better.
Metric: Final Resolution Rate (%): share of pdes disambiguation cases (requests with missing information (boundary conditions, material properties, geometry, solver settings or target outputs removed)) whose final specification resolves every annotated issue, the assistant may ask one clarification question per turn of a simulated scientist (Claude Sonnet 4.6, answering only from the hidden reference specification) for at most 11 turns before producing the final task specification, guided scientist-mode system prompt, judged by Gemini 2.5 Pro against the annotated issues; higher is better. Source: arxiv.org. Saturation forecast: Around February 2028. 5 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 78.7 | #185 |
| 2 | Gemini 2.5 Pro | 72.1 | #276 |
| 3 | GPT-OSS-120B | 65.6 | #704 |
| 4 | Gemini 2.5 Flash | 63.9 | #463 |
| 5 | Claude Sonnet 4.6 | 27.9 | #173 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=sciconvbench-disambiguation-pdes-final-resolution · How It Works · Data refreshed daily, snapshot 2026-11-21.