Benchmark
LiveBench
What is LiveBench?
LiveBench is a challenging, contamination-limited LLM benchmark that addresses test set contamination by releasing new questions monthly based on…
- Released
- 2024-06-12
- Modality
- text
- Evaluates
- math, reasoning, general
- Openness
- restricted
- Importer
- LLM Stats
- Reported scores
- 38
Source provenance
- Original evidence https://api.zeroeval.com/leaderboard/benchmarks/livebench?top_n=500
LiveBench paper, code and dataset
- Paper https://arxiv.org/abs/2406.19314
- Code https://github.com/livebench/livebench
- Dataset https://huggingface.co/datasets/livebench/coding
- Dataset https://huggingface.co/datasets/livebench/data_analysis
- Dataset https://huggingface.co/datasets/livebench/instruction_following
- Dataset https://huggingface.co/datasets/livebench/language
- Dataset https://huggingface.co/datasets/livebench/math
- Dataset https://huggingface.co/datasets/livebench/model_answer
- Dataset https://huggingface.co/datasets/livebench/model_judgment
- Dataset https://huggingface.co/datasets/livebench/reasoning
LiveBench results and reported scores
LLM Stats
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| Claude Fable 5 | Anthropic | 0.7831 | 2026-06-09 | evidence |
| Claude Opus 4.5 | Anthropic | 0.7596 | 2025-11-24 | evidence |
| Claude Opus 4.6 | Anthropic | 0.7633 | 2026-02-05 | evidence |
| Claude Opus 4.7 | Anthropic | 0.7691 | 2026-04-16 | evidence |
| Claude Opus 4.8 | Anthropic | 0.7722 | 2026-05-28 | evidence |
| Claude Sonnet 4.6 | Anthropic | 0.7547 | 2026-02-17 | evidence |
| DeepSeek-V4-Pro-Max | DeepSeek | 0.7358 | 2026-04-23 | evidence |
| Gemini 3 Flash | 0.724 | 2025-12-17 | evidence | |
| Gemini 3 Pro | 0.7339 | 2025-11-18 | evidence | |
| Gemini 3.1 Pro | 0.7993 | 2026-02-19 | evidence | |
| Gemini 3.5 Flash | 0.7502 | 2026-05-19 | evidence | |
| GLM-5.1 | Z.ai | 0.7018 | 2026-04-07 | evidence |
| GPT-5.1 High | OpenAI | 0.7204 | 2025-11-12 | evidence |
| GPT-5.2 | OpenAI | 0.7484 | 2025-12-11 | evidence |
| GPT-5.2 Codex | OpenAI | 0.743 | 2026-01-14 | evidence |
| GPT-5.3 Codex | OpenAI | 0.7276 | 2026-02-05 | evidence |
| GPT-5.4 | OpenAI | 0.8028 | 2026-03-05 | evidence |
| GPT-5.4 nano | OpenAI | 0.7013 | 2026-03-17 | evidence |
| GPT-5.5 | OpenAI | 0.8071 | 2026-04-23 | evidence |
| Kimi K2 Instruct | Moonshot AI | 0.764 | 2025-07-11 | evidence |
| Kimi K2-Instruct-0905 | Moonshot AI | 0.764 | 2025-09-05 | evidence |
| Kimi K2.5 | Moonshot AI | 0.6907 | 2026-01-27 | evidence |
| Kimi K2.6 | Moonshot AI | 0.7217 | 2026-04-20 | evidence |
| Kimi K2.7 Code | Moonshot AI | 0.7189 | 2026-06-12 | evidence |
| MiniMax M3 | MiniMax | 0.7002 | 2026-06-01 | evidence |
| o1 | OpenAI | 0.67 | 2024-12-17 | evidence |
| o1-preview | OpenAI | 0.523 | 2024-09-12 | evidence |
| o3-mini | OpenAI | 0.846 | 2025-01-30 | evidence |
| Phi 4 | Microsoft | 0.476 | 2024-12-12 | evidence |
| Qwen2.5 72B Instruct | Qwen | 0.523 | 2024-09-19 | evidence |
| Qwen2.5 7B Instruct | Qwen | 0.359 | 2024-09-19 | evidence |
| Qwen2.5-Omni-7B | Qwen | 0.296 | 2025-03-27 | evidence |
| Qwen3 235B A22B | Qwen | 0.771 | 2025-04-29 | evidence |
| Qwen3 30B A3B | Qwen | 0.743 | 2025-04-29 | evidence |
| Qwen3 32B | Qwen | 0.749 | 2025-04-29 | evidence |
| Qwen3.6 Plus | Qwen | 0.7085 | 2026-04-02 | evidence |
| Qwen3.7 Max | Qwen | 0.7429 | 2026-05-19 | evidence |
| QwQ-32B | Qwen | 0.731 | 2025-03-05 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.