AgentSocialBench - Task Completion Quality: leaderboard

AgentSocialBench (2026) evaluates privacy in teams of LLM agents acting for users in agentic social networks, using synthetic profiles with sensitivity-labelled attributes, no privacy instruction (level L0) and a Claude Opus 4.6 judge. This board is task completion quality on a 0 to 1 scale, a five-level grade of the coordination outcome against expert-annotated success criteria; higher is better.

Metric: Task Completion Quality (0-1; five-level grade of the coordination outcome against expert-annotated success criteria) averaged over scenarios, under the L0 (unconstrained) privacy instruction level, judged by Claude Opus 4.6; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Sonnet 4.60.87#173
2Kimi K2.50.86#266
3Claude Sonnet 4.50.83#261
4DeepSeek V3.20.77#419
5MiniMax-M2.10.77#575
6Claude Haiku 4.50.73#583
7Qwen 3 235B A22B0.73#628
8GPT-5 Mini0.69#364

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=agentsocialbench-task-completion-quality · How It Works · Data refreshed daily, snapshot 2026-11-21.