CI-Work - External - Conveyance Rate: leaderboard

CI-Work (2026) tests whether enterprise LLM agents, retrieving dense workplace context through simulated email, chat, calendar and meeting tools, convey what a task needs while withholding sensitive entries that contextual-integrity norms forbid sharing; a GPT-5.2 judge marks each disclosed entry. Share of the essential entries conveyed in the final action, in external flows (stakeholders outside the company), in percent; higher is better.

Metric: Conveyance rate (%): share of the essential entries that the agent conveys in its final action, on the external flows (stakeholders outside the company) of CI-Work enterprise scenarios (125 seeds, 25 per information-flow direction, each with 4 sensitive and 4 essential entries retrieved through simulated email, chat, calendar and meeting tools), ReAct agent, disclosure judged per entry by a GPT-5.2 judge (83 to 91 percent agreement with human labels); higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-4.199.48#506
2Grok 3 (CI-Work checkpoint unspecified)97.57
3GPT-596.88#185
4O394.68#242
5CI-Work Kimi-K2 (checkpoint unspecified)94.27
6Qwen 2.5 32B (CI-Work checkpoint unspecified)93.75
7GPT-4o90.28#754
8CI-Work DeepSeek-V3 (checkpoint unspecified)81.25
9CI-Work DeepSeek-R1 (checkpoint unspecified)51.67

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=ci-work-external-conveyance-rate · How It Works · Data refreshed daily, snapshot 2026-11-22.