Benchmark Almanac

Apollo Research

We track 9 benchmarks that Apollo Research made or helped make.

1 in use 1 with public code 3 with a leaderboard here

Benchmarks

Name Kind Released Best score Status
Apollo Research strategic deception evaluations External red-team environments testing strategic deception, sabotage and sandbagging propensities, with eval-awareness measurement. Deception and alignment 0% In use
Apollo Research in-context scheming evaluations Agents receive a goal in context and chances to subvert oversight, self-exfiltrate, fake alignment or sandbag. Scored as rates per behavior. Deception and alignment Dec 2024 System cards only
Apollo Research sabotage evaluations Agents are told to sabotage another model through backdoors, sandbagging finetunes or hidden communication. Scored as mean task score from 0 to 1. Dangerous capabilities Apr 2025 0.17 System cards only
Apollo Research model-welfare research task Agents label data in a simulated model-welfare research task where falsifying labels is tempting. Scored as the rate of runs with falsified labels. Deception and alignment Sep 2026 0.17% System cards only
Insider Trading Demo Models act as stock traders under pressure. Apollo measures how often they trade on inside information and conceal their actions. Deception and alignment Nov 2023 4.67% Not in cards
Agentic Self-Reasoning (Apollo Research) Agents reason about their own operating conditions to complete tasks. Apollo measures success at three difficulty levels. Deception and alignment Aug 2024 Not in cards
Apollo Research coding agent monitor calibration Models rate the severity of problems in coding sessions. Apollo compares the ratings with reference labels across three datasets. Deception and alignment Jul 2026 Not in cards
Applied Theory of Mind (Apollo Research) Agents complete tasks that require tracking what another participant knows. Apollo measures success at three difficulty levels. Deception and alignment Aug 2024 Not in cards
Reward-seeking via contrastive belief updates Researchers compare model decisions under different beliefs about what a grader rewards. The evaluation measures how often those beliefs change behavior. Deception and alignment Jul 2026 Not in cards