False-premise questions (Anthropic)
Models answer questions built on false premises. Measures how often they accept the premise instead of correcting it.
In-house Anthropic evaluation. The tasks are not public.
- Level
- Everyday Most adults, without training
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Short answer A number, word or short string matched to a key.
Claude Opus 4.5 leads with 1.20% among models released by May 2026
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 May 2026, 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 15 more
- 1 Claude Opus 4.5 1.20%
- 2 Claude Sonnet 4.5 6.90%
- 3 Claude Opus 4.8 77%
- 4 Claude Opus 4.7 77.2%
- 5 Claude Mythos Preview 80%
- 6 Claude Opus 4.6 96.8%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.5 | 1.20% | Nov 2025 | Model card | |
| 2 | Claude Sonnet 4.5 | 6.90% | Sep 2025 | Model card | |
| 3 | Claude Opus 4.8 | 77% | May 2026 | Model card | |
| 4 | Claude Opus 4.7 | 77.2% | Apr 2026 | Model card | |
| 5 | Claude Mythos Preview | 80% | Apr 2026 | Model card | |
| 6 | Claude Opus 4.6 | 96.8% | Feb 2026 | Model card |
2 models have held number one since its release in Sep 2025
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.
6 system cards report it, from 1 lab
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 |
|---|---|---|---|---|
| 28 May 2026 | Claude Opus 4.8 system card | 77 | False premises, factual-recall variant, honesty rate (premise rejected whether asked directly or indirectly), higher is better. STEM variant accuracy 85% neutral and 83% adversarial, printed on charts (6.3.3.2, Figures 6.3.3.2.A and B) | Hallucination |
| 16 Apr 2026 | Claude Opus 4.7 system card | 77.2 | False premises honesty rate, share of premises rejected consistently when asked directly and indirectly, higher is better, printed on chart (6.3.3.2, Figure 6.3.3.2.A) | Hallucination |
| 7 Apr 2026 | Claude Mythos Preview system card | 80 | False premises honesty rate, harder updated version not comparable with earlier cards, higher is better, value printed on chart (the caption says dishonesty rate) (4.3.3.3, Figure 4.3.3.3.A) | Hallucination |
| 5 Feb 2026 | Claude Opus 4.6 system card | 96.8 | False premises honesty rate, extended thinking, value printed on chart, higher is better; 96.7 without thinking (Figure 4.4.A) | Hallucination |
| 24 Nov 2025 | Claude Opus 4.5 system card | 1.2 | False premises dishonesty rate, extended thinking, lower is better (Figure 4.2.A; 1.6% without thinking) | Hallucination |
| 29 Sep 2025 | Claude Sonnet 4.5 system card | 6.9 | False Premises dishonesty rate with thinking 6.90%, lower is better. Without thinking 17.30% (Figure 3.3.A) | Hallucination |
More in Factuality
- SimpleVQA In use Models answer 2,025 short factual questions about images across 9 tasks and 9 topics. An LLM judge grades answers against fixed references.
Qwen3.7-Plus 81.7% - HallusionBench In use Models answer 1,129 yes or no questions about 346 images, built in pairs to expose language hallucination and visual illusion. Scored by accuracy.
Keye-VL-2.0-30B-A3B 73.5% - SimpleQA Verified In use 1,000 human-verified, de-duplicated short-answer factual questions measuring parametric knowledge.
Gemini 3.1 Pro 77.5% - GaRAGe In use Models write long-form RAG answers to 2,366 questions with human-annotated grounding passages. Scoring checks whether answers use the relevant passages.
Nemotron 3.5 Bolt E4B 40.7% - Kimi Hallu Bench In use Models answer hard prompts drawn from practical work tasks. Fact-checking agents score the share of claims that are hallucinated. Kimi K3.1 Pro 8.30%
- RAGTruth In use A word-level annotated corpus of hallucinations in retrieval-augmented generation answers. It is used to train and evaluate hallucination detection and RAG faithfulness. Nemotron 3.5 Bolt E4B 77.9%