DeepSWE
Agentic software engineering
v1.1codingactive
Score in %
Definition
- Version
- v1.1
- Category
- coding
- Maintainer
- not in the notes
- Status
- active
- Home page
- not in the notes
Sources
- OpenAI Launches GPT-6 Astra, Pricing, Benchmarks and Who Gets It First pasqualepillitteri.it https://pasqualepillitteri.it/en/news/14246/openai-launches-gpt-6-astra-pricing-benchmarks
- Cognition SWE-2 : Modèle, Harnais et les chiffres qui se cachent derrière orcarouter.ai https://www.orcarouter.ai/fr/blog/swe-2
- deepseek-ai/DeepSeek-V4.1-Flash , Hugging Face huggingface.co https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash
Scores at a Glance
Score in %
Higher is better for this unit.
1
Gemini 4 ArgonGoogle DeepMindGoogle self-reported
DeepSeek-V4.1-FlashDeepSeekmini-SWE
GPT-6 AstraOpenAI
SWE-2Cognition (owns Windsurf)Gemini 3.8 FlashGoogle DeepMindabove 70% (chart only)
GLM-5.3Z.ai (Zhipu)mini-swe-agent
Gemini 4 ArgonGoogle DeepMindGoogle self-reported77.9 %
vendor-reported2DeepSeek-V4.1-FlashDeepSeekmini-SWE74.2 %
vendor-reported3GPT-6 AstraOpenAI74.1 %
vendor-reported4SWE-2Cognition (owns Windsurf)73 %
vendor-reported5Gemini 3.8 FlashGoogle DeepMindabove 70% (chart only)70 %
vendor-reported6GLM-5.3Z.ai (Zhipu)mini-swe-agent66.9 %
vendor-reported0255075100
Where this comes from
Each bar is one model, with its most independent recorded score. Hatched bars are vendor-reported; the word beside each bar says who ran it. Settings and dates differ between rows; the Scores table below has every result.
Scores
| Model | Score | Who ran it | Setting | View or population | Date | Source |
|---|---|---|---|---|---|---|
| Gemini 4 Argon | 77.9 % | vendor-reported | Google self-reported | – | – | orcarouter.ai |
| DeepSeek-V4.1-Flash | 74.2 % | vendor-reported | mini-SWE | – | – | huggingface.co |
| GPT-6 Astra | 74.1 % | vendor-reported | – | – | – | pasqualepillitteri.it |
| SWE-2 | 73 % | vendor-reported | – | – | – | orcarouter.ai |
| Gemini 3.8 Flash | 70 % | vendor-reported | above 70% (chart only) | – | – | deepmind.google |
 MiMo-V2.6-Flash | 67.9 % | vendor-reported | – | MiMo-V2.6 model card comparison table | 21 Sep 2026 | huggingface.co |
| GLM-5.3 | 66.9 % | vendor-reported | mini-swe-agent | – | – | huggingface.co |
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "DeepSWE", https://aurion.semai.id/benchmarks/deepswe-v1-1/, records of October 2026.