Skip to main content
BenchLM

FrontierScience

We show this table for reference; we do not rank on it.

Data verified 34 confirmed releases in the last 30 daysFollow model changes

A benchmark for research-level scientific reasoning, designed to separate frontier models on difficult science tasks that mix domain knowledge with deep reasoning.

Benchmark score on FrontierScience — September 27, 2026

We compile the FrontierScience rows from provider self-reports. GPT-5.4 Pro leads the table at 36.7%. We do not use these results to rank models overall.

1 modelKnowledgeCurrentDisplay onlyUpdated September 27, 2026

Benchmark score table (1 model)

Score
1
GPT-5.4 ProOpenAI · Closed
36.7%

About FrontierScience

Year

2026

Tasks

Research-level science tasks

Format

Scientific reasoning benchmark

Difficulty

Research frontier

FrontierScience matters because GPQA-style knowledge alone is not enough for scientific copilots. It better reflects the kind of reasoning needed for research assistance and frontier technical work.

Freshness and provenance

Version

FrontierScience 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does FrontierScience measure?

A benchmark for research-level scientific reasoning, designed to separate frontier models on difficult science tasks that mix domain knowledge with deep reasoning.

Which model scores highest on FrontierScience?

GPT-5.4 Pro by OpenAI currently leads with a score of 36.7% on FrontierScience.

How many models are evaluated on FrontierScience?

1 AI models have been evaluated on FrontierScience on BenchLM.

Last updated: September 27, 2026 · BenchLM version FrontierScience 2026

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.