header-langage
简体中文
繁體中文
English
Tiếng Việt
한국어
日本語
ภาษาไทย
Türkçe
Scan to Download the APP
Partnered With:

Kimi K3 scored 69% in DeepSWE, only one percentage point behind Claude Fable 5

According to Vision One monitoring, the AI data company Datacurve has integrated Kimi K3 into the DeepSWE long-range software engineering benchmark. This benchmark tests whether a model can independently complete real, complex, long-term development tasks.

Kimi K3 achieved a task success rate of 69%. GPT-5.6 Sol and Claude Fable 5 scored 73% and 70%, respectively. Datacurve stated that Kimi K3 is the first open-weight model to reach the cutting-edge closed-source model level in DeepSWE.

DeepSWE consists of 113 original tasks from 91 active open-source projects, covering 5 programming languages. All models use the same set of tools. Tasks are also not directly adapted from existing code commits to reduce training data contamination.

The average cost per task for Kimi K3 is $4.65. GPT-5.6 Sol costs $8.39, and Claude Fable 5 costs $21.63.

举报 Correction/Report
RelatedRecommended
Citigroup and Bank of America Securities believe that while the Kimi K3 reduces inference costs, it may paradoxically stimulate more invocations, longer contexts, and agent-based tasks through the "Jevons Paradox," thereby driving up demand for GPU, HBM, DDR5, eSSD, and network infrastructure.
From Writing Code to Reviewing Code, AI is Beginning to Enter the Engineering Audit Scene
Correction/Report
Submit
Add Library
Visible to myself only
Public
Save
Choose Library
Add Library
Cancel
Finish