Today for AI
HOT RADAR
agentHEAT 11.9°

AI CURATED HOT STORY · 11/8/2026

Alibaba-Backed Benchmark Finds Coding Agents Fully Complete Only One in Four Real Tasks

1 reports archived1 independent sourcesupdated 11/8/2026, 17:50:59
Synthesis & Latest Updates
1 Sources Cross-Validated

Researchers from Alibaba and several Asian universities built Co-Design Bench from real user change requests, requiring coding agents to rework existing web apps with multimodal inputs such as images, audio, video and documents. The best of ten tested models, GPT-6 Astra, met only about 70 percent of requirements on average and fully completed roughly one in four tasks, sharply contrasting with near-perfect scores on older benchmarks.

LATEST/Researchers from Alibaba and several Asian universities built Co-Design Bench from real user change requests, requiring coding agents to rework existing web apps with multimodal inputs such as images, audio, video and documents. The best of ten tested models, GPT-6 Astra, met only about 70 percent of requirements on average and fully completed roughly one in four tasks, sharply contrasting with near-perfect scores on older benchmarks.

Hourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first
  1. The DecoderT2·78 pts

    Alibaba-Backed Benchmark Finds Coding Agents Fully Complete Only One in Four Real Tasks

    • Co-Design Bench is built from real user change requests, asking agents to modify existing web apps rather than generate code from scratch
    • The best of ten models, GPT-6 Astra, met about 70 percent of requirements on average but fully completed only roughly one in four tasks
    • Tasks include image, audio, video and document inputs, testing multimodal understanding and real engineering execution