AA-Omniscience: non-hallucination
How often a model declines to answer rather than inventing one, on questions it does not know. The counterweight to the accuracy score.
As of 22 Sept 2026, MiniCPM5-1B leads AA-Omniscience: non-hallucination on BenchLeader with 99.1%, ahead of G9v3-3B at 88.3%, across 501 model configurations with a published result.
- Published by
- Artificial Analysis
- Category
- Knowledge
- Index weight
- Reference only
- Models
- 501
- Data as of
- 22 Sept 2026
Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.
- 1MiniCPM5-1B (none)99.1%
- 2G9v3-3B88.3%
- 3G9v3-39A5B87.0%
- 4Command A+85.8%
- 5LFM2.5-2.6B84.0%
- 6Grok 4.3 (medium)83.1%
- 7MiniCPM5-1B (thinking)83.0%
- 8Grok 4.3 (low)82.6%
- 9Grok 4.20 (thinking)82.6%
- 10Qwen3.8 27B (none)81.9%
- 11MiniMax M381.6%
- 12Quasar 438B78.6%
- 13MiniCPM5-2B78.1%
- 14Ling 3.0 Flash VL78.0%
- 15K Exaone 2.0 080377.4%
501 of 501
| # | |||
|---|---|---|---|
| 1 |  MiniCPM5-1Bno reasoningOpenBMBopen | 99.1% | 43.8 |
| 2 | G9v3-3BAI9Starsopen | 88.3% | – |
| 3 | G9v3-39A5BAI9Starsopen | 87.0% | 52.8 |
| 4 |  Command A+Cohereopen | 85.8% | 51.1 |
| 5 |  LFM2.5-2.6BLiquid AIopen ↗ | 84.0% | 44.0 |
| 6 |  Grok 4.3mediumSpaceXAI | 83.1% | 59.5 |
| 7 | MiniCPM5-1BthinkingOpenBMBopen | 83.0% | 44.5 |
| 8 | Grok 4.3lowSpaceXAI | 82.6% | 57.6 |
| 9 | Grok 4.20thinkingSpaceXAI | 82.6% | 59.7 |
| 10 |  Qwen3.8 27Bno reasoningAlibabaopen ↗ | 81.9% | 50.6 |
| 11 |  MiniMax M3MiniMaxopen ↗ | 81.6% | 57.2 |
| 12 |  Quasar 438BMultiverse Computing | 78.6% | 56.8 |
| 13 | MiniCPM5-2BOpenBMBopen | 78.1% | 49.9 |
| 14 |  Ling 3.0 Flash VLAnt Groupopen ↗ | 78.0% | 56.4 |
| 15 |  K Exaone 2.0 0803LG AI Researchopen | 77.4% | 50.3 |
| 16 | Grok 4.6mediumSpaceXAI | 76.0% | 65.5 |
| 17 | Grok 4.6xhighSpaceXAI | 76.0% | 64.1 |
| 18 |  Solar Pro 4Upstage | 75.6% | 55.2 |
| 19 |  MiMo-V2.5-ProXiaomiopen ↗ | 75.3% | 58.9 |
| 20 | Solar Open2 250BUpstageopen ↗ | 74.6% | 54.5 |
| 21 |  Granite 4.2 30BIBMopen ↗ | 74.4% | 49.7 |
| 22 | Qwen3 7maxAlibaba | 74.4% | 60.8 |
| 23 |  Claude Haiku 4.5no reasoningAnthropic | 74.3% | 49.1 |
| 24 | Grok 4.3highSpaceXAI | 74.2% | 57.9 |
| 25 | Grok 3 minithinkingSpaceXAI | 74.2% | 50.0 |
| 26 |  K2 Horizon 375B A23BMBZUAIopen | 74.2% | 57.2 |
| 27 |  GLM 5.2maxZhipu AIopen ↗ | 73.7% | 63.6 |
| 28 | Granite 4.2 3BIBMopen ↗ | 73.7% | 45.5 |
| 29 | Claude Haiku 4.5thinkingAnthropic | 72.7% | 47.3 |
| 30 | GLM 5.3 FlashZhipu AIopen ↗ | 72.4% | 63.6 |
| 31 | Qwen3.7 PlusAlibaba | 72.3% | 59.5 |
| 32 |  Motif 3Motif Technologiesopen | 71.6% | 58.7 |
| 33 | Qwen3.8 Max (0902)maxAlibaba | 71.2% | 66.2 |
| 34 | Claude Sonnet 4thinkingAnthropic | 70.9% | 53.8 |
| 35 | NewGrok 4.7xhighSpaceXAI | 70.7% | 61.9 |
| 36 | GLM 5.3maxZhipu AIopen ↗ | 70.5% | 65.4 |
| 37 |  Nemotron 3 Ultra 550B A55BthinkingNVIDIAopen ↗ | 70.3% | 57.3 |
| 38 | GLM 5.1thinkingZhipu AIopen ↗ | 70.0% | 57.4 |
| 39 | Mimo v2 ProXiaomi | 70.0% | 59.6 |
| 40 | Qwen3.8 27BxhighAlibabaopen ↗ | 69.7% | 58.9 |
| 41 |  Gemma 3 270MGoogleopen ↗ | 69.6% | 39.3 |
| 42 | Ling 3.0 TinyAnt Groupopen | 69.5% | 49.9 |
| 43 | K2 Horizon MoVA 36B A4BMBZUAIopen | 69.5% | 54.6 |
| 44 | Grok 4.6lowSpaceXAI | 69.3% | 60.5 |
| 45 | Gemma 4 E4BthinkingGoogleopen ↗ | 69.1% | 45.0 |
| 46 |  Muse Spark 1.3xhighMeta | 68.5% | 67.9 |
| 47 | MiMo-V2.5Xiaomiopen ↗ | 68.1% | 55.5 |
| 48 | Granite 4.2 8BIBMopen ↗ | 68.0% | 46.7 |
| 49 | Gemma 4 E2BthinkingGoogleopen ↗ | 67.6% | 42.1 |
| 50 | NewGrok 4.7highSpaceXAI | 67.6% | – |
| 51 |  Agnes 2.5 Pro BetaSapiens AI | 67.2% | 60.3 |
| 52 | Muse Spark 1.3maxMeta | 67.1% | 69.2 |
| 53 |  A.X-K2SK Telecomopen | 67.0% | 53.7 |
| 54 | Muse Spark 1.2xhighMeta | 66.7% | 63.7 |
| 55 | K2 Horizon 7BMBZUAIopen | 66.4% | 52.4 |
| 56 | GLM 5.2no reasoningZhipu AIopen ↗ | 66.3% | 49.3 |
| 57 | Grok 4.6highSpaceXAI | 65.7% | 64.2 |
| 58 | Gemini 3.5 Flash LiteGoogle | 65.6% | 55.2 |
| 59 | Qwen3.6 PlusAlibaba | 65.4% | 58.4 |
| 60 | GLM-5thinkingZhipu AIopen ↗ | 64.7% | 59.4 |
Cite as: BenchLeader, “AA-Omniscience: non-hallucination leaderboard”, https://www.benchleader.com/benchmarks/aa_omniscience_non_hallucination, data as of 22 Sept 2026.
AA-Omniscience: non-hallucination: questions
- What does AA-Omniscience: non-hallucination measure?
- How often a model declines to answer rather than inventing one, on questions it does not know. The counterweight to the accuracy score. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads AA-Omniscience: non-hallucination?
- MiniCPM5-1B leads AA-Omniscience: non-hallucination with 99.1% as of 22 Sept 2026, ahead of G9v3-3B at 88.3%.
- How many models have AA-Omniscience: non-hallucination results?
- 501 model configurations have a AA-Omniscience: non-hallucination result on BenchLeader, all taken from Artificial Analysis.
- Who runs AA-Omniscience: non-hallucination and how often is it updated?
- AA-Omniscience: non-hallucination is published by Artificial Analysis. BenchLeader re-reads the published results every morning and records the date each result was published.
- Does AA-Omniscience: non-hallucination count toward the BenchLeader Index?
- No. AA-Omniscience: non-hallucination is shown for reference but left out of the composite index.