# FrontierSWE v2

> A 34-task expansion of FrontierSWE for ultra-long-horizon engineering and research work that remains far from saturation.

Canonical page: https://benchlm.ai/benchmarks/frontierswev2

- Category: [Coding](/coding)
- Last updated: October 15, 2026

## About FrontierSWE v2

- Year: 2026
- Tasks: 34 ultra-long-horizon engineering and research tasks
- Format: Five-trial mean task score (Mean@5), 0-100
- Difficulty: Ultra-long-horizon frontier software engineering
- Paper: [FrontierSWE v2](https://www.frontierswe.com/blog/v2)

FrontierSWE v2 keeps 13 of the original 17 tasks, retires four as saturated or unreliably scorable, and adds 21 new tasks across visual reasoning, scientific computing, and AI research. Proximal runs every model at maximum reasoning effort in its own Proximus harness with a 20-hour budget and five trials per task, and publishes mean, best, and worst scores on a 0-100 scale. BenchLM mirrors the public Mean@5 table as benchmark-owned rows; models that only appear in provider material stay provider-exact on the same scale.

The mirrored table is for reference; the scores BenchLM stores for each model feed the ranking. BenchAlign v5.8 gives FrontierSWE v2 8% of the Coding reference weight, so it moves the Coding leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.

## Leaderboard (21 models)

| Rank | Model | Configuration | Creator | Score |
|------|-------|---------------|---------|-------|
| 1 | [GPT-6 Astra](/models/gpt-6-astra) | proximus | OpenAI | 65.5% |
| 2 | [Claude Opus 5.5](/models/claude-opus-5-5) | proximus | Anthropic | 62.3% |
| 3 | [Claude Fable 5.1](/models/claude-fable-5-1) | proximus | Anthropic | 56.3% |
| 4 | [Gemini 4 Argon](/models/gemini-4-argon) | proximus | Google | 55.0% |
| 5 | [Claude Opus 5](/models/claude-opus-5) | proximus | Anthropic | 52.0% |
| 6 | [Claude Fable 5](/models/claude-fable) | proximus | Anthropic | 47.0% |
| 7 | [Claude Haiku 5.5](/models/claude-haiku-5-5) | proximus | Anthropic | 43.8% |
| 8 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | proximus | OpenAI | 32.2% |
| 9 | [GLM-5.3](/models/glm-5-3) | proximus | Z.AI | 30.2% |
| 10 | [Grok 4.7](/models/grok-4-7) | proximus | xAI | 29.5% |
| 11 | [Kimi K3](/models/kimi-k3) | proximus | Moonshot AI | 25.9% |
| 12 | [Muse Spark 1.3](/models/muse-spark-1-3) | proximus | Meta | 25.8% |
| 13 | [Grok 4.6](/models/grok-4-6) | proximus | xAI | 25.3% |
| 14 | [Gemini 3.7 Flash](/models/gemini-3-7-flash) | proximus | Google | 20.3% |
| 15 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | proximus | Google | 19.6% |
| 16 | [GLM-5.3-Flash](/models/glm-5-3-flash) | proximus | Z.AI | 18.1% |
| 17 | [Qwen3.8-Max-0902](https://www.frontierswe.com/) | proximus | Alibaba | 17.8% |
| 18 | [Qwen3.8 Max](/models/qwen3-8-max) | proximus | Alibaba | 15.8% |
| 19 | [DeepSeek V4 Flash Vision Exp](https://www.frontierswe.com/) | proximus | DeepSeek | 14.8% |
| 20 | [Muse Spark 1.2](/models/muse-spark-1-2) | proximus | Meta | 12.0% |
| 21 | [Inkling](/models/inkling) | proximus | Thinking Machines Lab | 4.1% |

## FAQ

### What does FrontierSWE v2 measure?

A 34-task expansion of FrontierSWE for ultra-long-horizon engineering and research work that remains far from saturation.

### Which model leads the published FrontierSWE v2 snapshot?

GPT-6 Astra currently leads the published FrontierSWE v2 snapshot with a score of 65.5%.

### How many models are evaluated on FrontierSWE v2?

The October 15, 2026 contains 21 AI models.

### Does FrontierSWE v2 affect BenchLM's overall score?

Yes. BenchAlign v5.8 gives FrontierSWE v2 8% of the Coding reference weight, so it moves the Coding leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.