# FrontierScience Research

> A research-focused FrontierScience evaluation variant for scientific investigation and problem solving.

Canonical page: https://benchlm.ai/benchmarks/frontierscienceresearch

- Category: [Knowledge](/knowledge)
- Last updated: September 27, 2026

## About FrontierScience Research

- Year: 2026
- Tasks: Scientific research problems
- Format: Research evaluation
- Difficulty: Frontier scientific research
- Paper: [Muse Spark Eval Methodology](https://ai.meta.com/static-resource/muse-spark-eval-methodology)

Meta uses FrontierScience Research in its Contemplating-mode comparison table as a distinct scientific research variant. BenchLM stores it as a display-only frontier science reference.

FrontierScience Research is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (4 models)

| Rank | Model | Creator | Score |
|------|-------|---------|-------|
| 1 | [Apodex 1.1](/models/apodex-1-1) | Apodex | 63.3% |
| 2 | [Apodex 1.1 Mini](/models/apodex-1-1-mini) | Apodex | 51.7% |
| 3 | [GPT-5.4 Pro](/models/gpt-5-4-pro) | OpenAI | 36.7% |
| 4 | [Agents-A1-4B](/models/agents-a1-4b) | InternScience | 33.3% |

## FAQ

### What does FrontierScience Research measure?

A research-focused FrontierScience evaluation variant for scientific investigation and problem solving.

### Which model scores highest on FrontierScience Research?

Apodex 1.1 by Apodex currently leads with a score of 63.3% on FrontierScience Research.

### How many models are evaluated on FrontierScience Research?

4 AI models have been evaluated on FrontierScience Research on BenchLM.

### Does FrontierScience Research affect BenchLM's overall score?

Not directly. FrontierScience Research is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Compare Top Models on FrontierScience Research

- [Apodex 1.1 vs Apodex 1.1 Mini](/compare/apodex-1-1-vs-apodex-1-1-mini)
- [Apodex 1.1 Mini vs GPT-5.4 Pro](/compare/apodex-1-1-mini-vs-gpt-5-4-pro)
- [GPT-5.4 Pro vs Agents-A1-4B](/compare/agents-a1-4b-vs-gpt-5-4-pro)