Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-29 · source-backed
Alibaba International's Accio team open-sourced 107 tasks (53 CLI, 28 browser, 16 file, 10 API/MCP) running against fourteen offline replicas of real business software in a fresh container per task, with verifiers inspecting mock-service state rather than the transcript. Claude Opus 5 leads at 65/107 (60.7%) using 52.5 steps, 7.8 minutes and 2.05M tokens per task, ahead of Opus 4.8 at 56/107, with Qwen 3.8 Max and DeepSeek V4 Pro tied at 53/107. State-inspecting verifiers are the design decision every agent benchmark should copy. (GitHub)
Each link below shares sources, entities, or timing with this story.
RealReplicaBench built by Accio / Shared entities / Same source domain / Shared topic / Earlier coverage
Linked by a graph relationship (RealReplicaBench built by Accio); both cover Accio, Claude Opus, CLI, GitHub; reported by the same outlet (github.com).
Claude Code uses Opus / Shared entities / Same source domain / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (Claude Code uses Opus); both cover GitHub, MCP, Opus; reported by the same outlet (github.com).
DeepSeek V4 Pro benchmarked against CommerceAgentBench / Shared entities / Same source / Shared topic
Linked by a graph relationship (DeepSeek V4 Pro benchmarked against CommerceAgentBench); both cover Accio, CommerceAgentBench, DeepSeek V4 Pro; cite the same source (GitHub).
Cursor uses Opus / Shared entities / Same source domain / Shared topic / Earlier coverage
Linked by a graph relationship (Cursor uses Opus); both cover GitHub, MCP, Qwen; reported by the same outlet (github.com).
RealReplicaBench built by Accio / Shared entities / Same source domain / Shared topic / Earlier coverage
Linked by a graph relationship (RealReplicaBench built by Accio); both cover Accio, Claude Opus, CLI; reported by the same outlet (github.com).
DeepSeek V4 Pro benchmarked against CommerceAgentBench / Shared entities / Same source domain / Shared topic / Earlier coverage
Linked by a graph relationship (DeepSeek V4 Pro benchmarked against CommerceAgentBench); both cover DeepSeek V4 Pro, GitHub, Opus; reported by the same outlet (github.com).
Claude Code uses Opus / Shared entities / Same source domain / Shared topic / Earlier coverage
Linked by a graph relationship (Claude Code uses Opus); both cover Claude Opus, GitHub, Opus; reported by the same outlet (github.com).
Linked by a graph relationship (Claude Code uses Opus); both cover Claude Opus, Qwen; reported by the same outlet (github.com).