Alibaba's benchmark highlights challenges for AI coding agents

Agents & Dev✍ TextPublished: Source: The Decoder

#Alibaba Qwen #OpenAI

Key point

According to Alibaba's Co-Design Bench, even the top-performing GPT-6 Astra fully solved only 26.4% of tasks in real-world tests.

Alibaba has introduced Co-Design Bench, a new coding-agent benchmark that tests AI agents on real user requests that have previously caused them to fail in practice.

In the test, even the top performer, GPT-6 Astra, fully solved only 26.4% of the tasks. This contrasts with older benchmarks, where the same model scores above 99%.

This item is a summary. The full, checkable version is at the original source:

Read the original at The Decoder ↗