LiveBench 2.0 Reasoning
LiveBench 2.0 reasoning suite: short logic, spatial and theory-of-mind questions with one correct answer.
As of 21 Oct 2026, Gemini 4 Argon leads LiveBench 2.0 Reasoning on BenchLeader with 95.1%, ahead of GPT-6 Astra at 92.7%, across 48 model configurations with a published result.
- Published by
- LiveBench
- Category
- Reasoning
- Index weight
- Reference only
- Models
- 48
- Data as of
- 21 Oct 2026
LiveBench (livebench.ai), White et al. 2024.
Not comparable with the earlier test of the same name, whose final scores stay on LiveBench 1.0 Reasoning.
What the test looks like
LiveBench 2.0's reasoning suite: short logic, spatial and theory-of-mind questions, each with a single correct answer.
How it is scored
Percent correct on the current release, the equal-weight mean of its subcategories, graded automatically.
What to keep in mind
It shares a name with LiveBench 1.0's Reasoning category but is a different question set, so the two are kept apart; 2.0 scores start on 19 October 2026.
- 1Gemini 4 Argon95.1%
- 2GPT-6 Astra92.7%
- 3GPT-6.1 Sol92.6%
- 4Claude Opus 5.592.2%
- 5Claude Fable 5.191.7%
- 6GPT-5.6 Sol91.7%
- 7Claude Sonnet 5.591.6%
- 8Claude Opus 591.2%
- 9Grok 4.691.2%
- 10Kimi K390.7%
- 11Muse Wonder90.6%
- 12GPT-5.6 Terra90.6%
- 13Muse Spark 1.290.0%
- 14Claude Fable 589.7%
- 15GPT-5.589.7%
| # | ||||
|---|---|---|---|---|
| 1 |  Gemini 4 Argon | 95.1% | 63.3 | 2026-10-14 |
| 2 |  GPT-6 Astra | 92.7% | 67.9 | 2026-10-14 |
| 3 | GPT-6.1 Sol | 92.6% | 63.3 | 2026-10-14 |
| 4 |  Claude Opus 5.5 | 92.2% | 61.9 | 2026-10-14 |
| 5 | Claude Fable 5.1 | 91.7% | 66.2 | 2026-10-14 |
| 6 | GPT-5.6 Sol | 91.7% | 55.9 | 2026-10-14 |
| 7 | Claude Sonnet 5.5 | 91.6% | – | 2026-10-14 |
| 8 | Claude Opus 5 | 91.2% | 65.4 | 2026-10-14 |
| 9 |  Grok 4.6 | 91.2% | 57.8 | 2026-10-14 |
| 10 |  Kimi K3open ↗ | 90.7% | 59.1 | 2026-10-14 |
| 11 |  Muse Wonder | 90.6% | 63.6 | 2026-10-14 |
| 12 | GPT-5.6 Terra | 90.6% | 52.6 | 2026-10-14 |
| 13 | Muse Spark 1.2 | 90.0% | – | 2026-10-14 |
| 14 | Claude Fable 5 | 89.7% | 67.1 | 2026-10-14 |
| 15 | GPT-5.5 | 89.7% | 60.2 | 2026-10-14 |
| 16 | Muse Spark 1.3 | 89.7% | 58.3 | 2026-10-14 |
| 17 | Gemini 3.8 Flash | 89.3% | 55.6 | 2026-10-14 |
| 18 | Claude Opus 4.8 | 89.2% | 60.1 | 2026-10-14 |
| 19 | New DeepSeek V4.1 Flash 1017open ↗ | 89.1% | – | 2026-10-14 |
| 20 | Claude Sonnet 5 | 88.7% | 56.0 | 2026-10-14 |
| 21 | GPT-6 Sol | 88.7% | – | 2026-10-14 |
| 22 |  Qwen3.8 Maxmax | 88.2% | 63.3 | 2026-10-14 |
| 23 | Gemini 3.7 Flash | 87.8% | 53.2 | 2026-10-14 |
| 24 | Qwen3.8 Flashopen ↗ | 87.4% | – | 2026-10-14 |
| 25 | Grok 4.5 | 87.2% | 58.9 | 2026-10-14 |
| 26 | DeepSeek V4.1 Flashopen ↗ | 86.7% | – | 2026-10-14 |
| 27 | Claude Sonnet 4.6 | 86.4% | 57.1 | 2026-10-14 |
| 28 |  GLM 5.3open ↗ | 85.8% | – | 2026-10-14 |
| 29 | GPT-5.6 Luna | 85.6% | 47.9 | 2026-10-14 |
| 30 | Deepseek v4 Flash Vision | 85.4% | – | 2026-10-14 |
| 31 | Gemini 3.1 Pro | 84.0% | 60.7 | 2026-10-14 |
| 32 |  Mistral Large 4open | 83.9% | 56.7 | 2026-10-14 |
| 33 |  Step 5 Previewopen | 83.8% | 60.6 | 2026-10-14 |
| 34 | New MiniMax-M3.1-Flash | 82.9% | 61.9 | 2026-10-14 |
| 35 | Grok 4.7 | 82.7% | – | 2026-10-14 |
| 36 | Claude Haiku 5.5 | 82.4% | – | 2026-10-14 |
| 37 | GPT-6 Luna | 81.8% | – | 2026-10-14 |
| 38 | New Beam 1.0open | 80.2% | 54.6 | 2026-10-14 |
| 39 | Qwen3.8 27Bopen ↗ | 80.0% | 54.5 | 2026-10-14 |
| 40 |  MiMo-V2.6-Proopen ↗ | 79.6% | 64.1 | 2026-10-14 |
| 41 |  Nemotron Super VL 3.5 | 78.8% | – | 2026-10-14 |
| 42 | GLM 5.3 Flashopen ↗ | 77.6% | 62.7 | 2026-10-14 |
| 43 | Nemotron Ultra 3 | 74.7% | – | 2026-10-14 |
| 44 | MiniMax M3open ↗ | 74.5% | 56.5 | 2026-10-14 |
| 45 | NewMuse Spark Openopen ↗ | 73.7% | – | 2026-10-14 |
| 46 | Gemma 4 31Bopen ↗ | 68.3% | 48.3 | 2026-10-14 |
| 47 | Claude Haiku 4.5 | 61.7% | 45.3 | 2026-10-14 |
| 48 | Nemotron Lightning 3.5 | 42.9% | – | 2026-10-14 |
More reasoning benchmarks · All LLM benchmarks
Cite as: BenchLeader, “LiveBench 2.0 Reasoning leaderboard”, https://www.benchleader.com/benchmarks/livebench2_reasoning, data as of 21 Oct 2026.
LiveBench 2.0 Reasoning: questions
- What does LiveBench 2.0 Reasoning measure?
- LiveBench 2.0's reasoning suite: short logic, spatial and theory-of-mind questions, each with a single correct answer. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads LiveBench 2.0 Reasoning?
- Gemini 4 Argon leads LiveBench 2.0 Reasoning with 95.1% as of 21 Oct 2026, ahead of GPT-6 Astra at 92.7%.
- How many models have LiveBench 2.0 Reasoning results?
- 48 model configurations have a LiveBench 2.0 Reasoning result on BenchLeader, all taken from LiveBench.
- Who runs LiveBench 2.0 Reasoning and how often is it updated?
- LiveBench 2.0 Reasoning is published by LiveBench. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does LiveBench 2.0 Reasoning count toward the BenchLeader Index?
- No. LiveBench 2.0 Reasoning is shown for reference but left out of the composite index.