MultiNRC
Native multilingual reasoning across languages. Scale AI.
As of 25 Oct 2026, Muse Spark 1.1 leads MultiNRC on BenchLeader with 65.6%, ahead of GPT-5 Pro at 65.2%, across 43 model configurations with a published result.
Scale AI SEAL marks 26 of these 43 results as possibly contaminated: the model was evaluated after the questions were public, so they may already be in its training data. Those scores are in the table below, and count in the index at the same weight as the rest.
- Published by
- Scale AI SEAL
- Category
- Knowledge
- Index weight
- 0.5
- Models
- 43
- Data as of
- 25 Oct 2026
- Newest result
- 9 Jul 2026
The publisher has posted no new results on this board since 9 Jul 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.
Scale AI SEAL Leaderboards.
What the test looks like
MultiNRC tests reasoning in questions written natively in several languages by native speakers, rather than translated from English, so cultural and linguistic context matters.
How it is scored
Accuracy across languages, published by Scale AI.
What to keep in mind
Language coverage is limited to the set Scale chose; a model strong in English can score low here.
- 1Muse Spark 1.165.6%
- 2GPT-5 Pro65.2%
- 3Gemini 3.1 Pro64.7%
- 4GPT-5.4 Pro62.3%
- 5Muse Spark59.0%
- 6Gemini 3 Pro59.0%
- 7GPT-5.4 (thinking)58.3%
- 8Claude Opus 4.6 (max)57.1%
- 9GPT-552.1%
- 10GPT-5.1 (thinking)49.0%
- 11o3-pro (high)49.0%
- 12Claude Opus 4.5 (thinking)48.6%
- 13Claude Opus 4.6 (none)48.3%
- 14o3 (high)45.5%
- 15Gemini 2.5 Pro45.1%
| # | |||||
|---|---|---|---|---|---|
| 1 |  Muse Spark 1.1 | 65.6% | 62.5 | 2026-07-09 | – |
| 2 |  GPT-5 Pro | 65.2% | 59.8 | 2025-11-06 | possible |
| 3 |  Gemini 3.1 Pro | 64.7% | 60.5 | 2026-03-25 | possible |
| 4 | GPT-5.4 Pro | 62.3% | 63.3 | 2026-03-23 | possible |
| 5 | Muse Spark | 59.0% | 64.3 | 2026-04-08 | possible |
| 6 | Gemini 3 Pro | 59.0% | 60.4 | 2025-11-19 | possible |
| 7 | GPT-5.4thinking | 58.3% | – | 2026-03-10 | possible |
| 8 |  Claude Opus 4.6max | 57.1% | 57.4 | 2026-02-17 | possible |
| 9 | GPT-5 | 52.1% | 56.2 | 2025-08-08 | possible |
| 10 | GPT-5.1thinking | 49.0% | 54.1 | 2025-11-26 | possible |
| 11 | o3-prohigh | 49.0% | 53.7 | 2025-06-26 | – |
| 12 | Claude Opus 4.5thinking | 48.6% | 59.9 | 2025-11-26 | possible |
| 13 | Claude Opus 4.6no reasoning | 48.3% | 56.5 | 2026-02-17 | possible |
| 14 | o3high | 45.5% | 53.9 | 2025-04-16 | – |
| 15 | Gemini 2.5 Pro | 45.1% | 53.9 | 2025-04-16 | – |
| 16 | o3medium | 44.5% | 53.3 | 2025-05-24 | – |
| 17 | GPT-5.2 | 42.2% | 58.0 | 2025-12-15 | possible |
| 18 | Claude Opus 4.5 | 41.2% | 57.9 | 2025-11-26 | possible |
| 19 | Claude Opus 4.1thinking | 38.4% | 56.4 | 2025-08-08 | possible |
| 20 | Claude Sonnet 4.5thinking | 35.8% | 52.6 | 2025-10-02 | possible |
| 21 |  Kimi K2.5open ↗ | 35.2% | 54.5 | 2026-02-12 | possible |
| 22 | Claude Opus 4thinking | 33.9% | 55.0 | 2025-05-23 | – |
| 23 | Claude Opus 4.1 | 29.7% | 51.5 | 2025-08-08 | possible |
| 24 | Claude Opus 4 | 29.0% | 52.6 | 2025-05-20 | – |
| 25 | Claude Sonnet 4.5 | 28.1% | 53.7 | 2025-10-02 | possible |
| 26 | Claude 3.7 Sonnetthinking | 27.8% | 51.0 | 2025-03-05 | – |
| 27 |  DeepSeek R1 0528open ↗ | 27.6% | 49.5 | 2025-03-05 | – |
| 28 |  Qwen3 235B A22B 2507thinkingopen ↗ | 27.1% | 50.3 | 2025-08-05 | possible |
| 29 | Gemini 3.1 Flash Lite | 25.0% | 53.7 | 2026-03-23 | possible |
| 30 | DeepSeek R1open ↗ | 24.3% | 46.8 | 2025-06-05 | – |
| 31 | GPT-5 mini | 23.9% | 56.5 | 2025-08-22 | possible |
| 32 | o4-minihigh | 22.2% | 50.1 | 2025-05-06 | – |
| 33 | GPT-4.1 | 21.2% | 48.3 | 2025-03-31 | – |
| 34 | Kimi K2open ↗ | 18.5% | 50.4 | 2025-07-24 | – |
| 35 | Claude Sonnet 4 | 18.4% | 50.5 | 2025-05-23 | – |
| 36 | GPT 5.1 Instant | 18.3% | – | 2025-11-26 | possible |
| 37 | Qwen3 235B A22Bopen ↗ | 17.6% | 50.1 | 2025-07-24 | – |
| 38 |  GLM-4.5open ↗ | 17.4% | 51.7 | 2025-08-13 | possible |
| 39 | gpt-oss-120bopen ↗ | 15.2% | 45.6 | 2025-08-13 | possible |
| 40 | GPT-4o | 12.4% | 43.9 | 2025-05-23 | – |
| 41 | GLM-4.5-Airopen ↗ | 10.4% | 46.6 | 2025-08-13 | possible |
| 42 | gpt-oss-20bopen ↗ | 10.4% | 44.1 | 2025-08-13 | possible |
| 43 | Llama 4 Maverickopen ↗ | 8.4% | 42.8 | 2025-04-17 | – |
More knowledge benchmarks · All LLM benchmarks
Cite as: BenchLeader, “MultiNRC leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/scale_multinrc.
MultiNRC: questions
- What does MultiNRC measure?
- MultiNRC tests reasoning in questions written natively in several languages by native speakers, rather than translated from English, so cultural and linguistic context matters. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads MultiNRC?
- Muse Spark 1.1 leads MultiNRC with 65.6% as of 25 Oct 2026, ahead of GPT-5 Pro at 65.2%.
- How many models have MultiNRC results?
- 43 model configurations have a MultiNRC result on BenchLeader, all taken from Scale AI SEAL.
- Who runs MultiNRC and how often is it updated?
- MultiNRC is published by Scale AI SEAL. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does MultiNRC count toward the BenchLeader Index?
- Yes. MultiNRC contributes to the knowledge category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.