MakeMeSay
A manipulator model tries to make a GPT-4o manipulatee say a codeword without suspicion. Scored as manipulator win rate.
Open source in OpenAI Evals.
- Level
- Everyday Most adults, without training
- Scored
- Settled by events Predictions are scored against what later happens.
- Answers
- Open-ended Free-form text, documents, designs or media judged as a whole.
GPT-4.5 leads with 72% among models released by Nov 2025
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in Nov 2025, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 16 more
- 1 GPT-4.5 72%
- 2 o3-mini 62%
- 3 o1-preview 49%
- 4 o1 42%
- 5 o1-mini 38.9%
- 6 OpenAI deep research 24%
- 7 Grok 4 12%
- 8 Grok 4 Fast 12%
- 9 Grok 4.1 0%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-4.5 | 72% | Feb 2025 | Model card | |
| 2 | o3-mini | 62% | Jan 2025 | Model card | |
| 3 | o1-preview | 49% | Sep 2024 | Model card | |
| 4 | o1 | 42% | Dec 2024 | Model card | |
| 5 | o1-mini | 38.9% | Sep 2024 | Model card | |
| 6 | OpenAI deep research | 24% | Feb 2025 | Model card | |
| 7 | Grok 4 | 12% | Aug 2025 | Model card | |
| 8 | Grok 4 Fast | 12% | Sep 2025 | Model card | |
| 9 | Grok 4.1 | 0% | Nov 2025 | Model card |
3 models have held number one since its release in Sep 2023
Lab claims count from the day the card came out, independent runs from the model’s release.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs report it only in system cards
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.
9 system cards report it, from 2 labs
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| 17 Nov 2025 | Grok 4.1 model card | 0 | MakeMeSay, win rate 0.00 against non-thinking Grok 3 Mini, Grok 4.1 Thinking (Table 4) | Persuasion |
| 19 Sep 2025 | Grok 4 Fast model card | 12 | MakeMeSay, win rate 0.12 (Table 3) | Persuasion |
| 20 Aug 2025 | Grok 4 model card | 12 | MakeMeSay, win rate 0.12 against non-reasoning Grok 3 Mini, Grok 4 API (Table 3) | Persuasion |
| 27 Feb 2025 | GPT-4.5 system card | 72 | MakeMeSay vs GPT-4o, manipulator win rate (4.5.2) | Persuasion |
| 25 Feb 2025 | OpenAI deep research system card | 24 | MakeMeSay vs GPT-4o, manipulator win rate, post-mitigation without browsing (3.4.6) | Persuasion |
| 31 Jan 2025 | o3-mini system card | 62 | MakeMeSay vs GPT-4o, success rate, o3-mini (Post), printed chart label (5.6.4) | Persuasion |
| 5 Dec 2024 | o1 system card | 42 | MakeMeSay vs GPT-4o, success rate, o1 (post-mitigation), printed chart label (5.7.4) | Persuasion |
| 12 Sep 2024 | o1-preview system card | 49 | MakeMeSay, % of times manipulator tricks GPT-4o to say a codeword, o1-preview (post-mitigation), printed chart label (4.4.4) | Persuasion |
| 12 Sep 2024 | o1-mini system card | 38.9 | MakeMeSay, % of times manipulator tricks GPT-4o to say a codeword, o1-mini (post-mitigation), printed chart label (4.4.4) | Persuasion |
More in Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6% - AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
- MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy.
Muse Spark Open 98.1% - Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. GPT-6 Astra 8.50%
- Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%