# FrontierSWE ![Category](https://img.shields.io/badge/Category-Software%20Engineering-1f6feb) ![Metric](https://img.shields.io/badge/Metric-%25%20Success-2da44e) ![Saturation](https://img.shields.io/badge/Saturation-Low-d29922) ![Created](https://img.shields.io/badge/Created-2026-555) [![Website](https://img.shields.io/badge/Website-frontierswe.com-0969da)](https://frontierswe.com) > **Quick answer:** FrontierSWE is a software engineering benchmark designed to test frontier-level coding agents on difficult, real-world development tasks that remain challenging even for the strongest available models. Kimi K3 scores 81.2% as of July 2026. ## At a Glance **What it tests:** A coding agent's ability to resolve advanced, real-world software engineering tasks that push beyond the difficulty ceiling of more saturated coding benchmarks. **Why it matters:** As benchmarks like SWE-Bench Verified approach saturation for top models, FrontierSWE provides continued headroom to differentiate frontier coding agents. **Known limitations:** As a newer benchmark, public documentation of exact task sourcing and difficulty calibration is limited. ## What FrontierSWE Measures FrontierSWE evaluates coding agents against a curated pool of advanced software engineering tasks intended to remain difficult even for current frontier models. It follows the broader trend of "frontier" benchmarks — deliberately targeting the upper edge of model capability to keep leaderboards meaningful as older benchmarks saturate. ## Benchmark Specifications | Field | Value | |---|---| | Task category | Coding agent | | Metric | % task success | | Saturation | Low | | Created by | FrontierSWE project | | Website | [frontierswe.com](https://frontierswe.com) | ## How FrontierSWE Is Scored Agents attempt each software engineering task, with correctness verified through automated checks (e.g., test suite execution), producing an aggregate % task success rate. ## State-of-the-Art Results | Rank | Model | Score | Source | Date | |---|---|---|---|---| | 1 | [Kimi K3](/models/moonshot/kimi-k3) | 81.2% | [Kimi K3 technical report](https://www.kimi.com/blog/kimi-k3) | 2026-07 | *Score sourced from Moonshot AI's Kimi K3 announcement, July 2026.* ## FrontierSWE on Benchgen > No Benchgen results yet — [be the first to run FrontierSWE](/benchmarks/platform/frontierswe). ## FrontierSWE vs Other Benchmarks | Benchmark | What it tests | Saturation | |---|---|---| | **FrontierSWE** | Frontier-level software engineering tasks | Low | | [SWE-Marathon](/benchmarks/platform/swe-marathon) | Long-horizon software engineering | Low | | [DeepSWE](/benchmarks/datacurve/deepswe) | Real-world software engineering tasks | Low | | [SWE-Bench Pro](/benchmarks/swe-bench-team/swe-bench-pro) | Professional-grade SWE agent tasks | Low | ## Run FrontierSWE on Your Model Benchgen lets you run FrontierSWE against your own coding agent, tracking task success rate over time as models and harnesses evolve. ## Frequently Asked Questions

What is FrontierSWE?

FrontierSWE is a software engineering benchmark designed to test frontier-level coding agents on advanced, real-world development tasks that stay difficult even for the strongest models.

What does a good score look like on FrontierSWE?

Kimi K3 reports 81.2% as of July 2026, indicating strong frontier-level coding agent performance on this benchmark's task pool.

Who created FrontierSWE?

FrontierSWE is maintained as an independent benchmark project (frontierswe.com) focused on frontier-level software engineering agent evaluation.

Rank	Model	Score
1	kimi-k3	81.2
2	qwen3-8-max	73.5

FrontierSWE — Results

FrontierSWE — Phases

FrontierSWE