Best LLMs for Knowledge — July 2026 Leaderboard
As of July 2026, the top knowledge model on the BenchLM leaderboard is Muse Spark 1.1 with a weighted knowledge score of 93.5.
Data refreshed:
General knowledge and factual understanding
Decision lens: use provisional-ranked mode for broader public evidence and verified-ranked mode for source-only comparisons. A model can move between views as evidence coverage changes.
- Data refreshed
- July 23, 2026
- Provisional-ranked
- 52 of 290 models
- Verified-ranked
- 27 of 290 models
- Weighted evidence
- 5 of 15 benchmarks
15 tracked benchmarks
MMLU, GPQA, GPQA-D, SuperGPQA, MMLU-Pro, HLE, FrontierScience, HLE w/o tools, SimpleQA, HealthBench Hard, HealthBench Professional, MedXpertQA (Text), FrontierScience Research, MMLU-Pro (Arcee), MMMLU
Scope: Frontier science, Broad academic knowledge, Factuality
Evidence set: MMLU, GPQA, GPQA-D, SuperGPQA, MMLU-Pro, HLE, FrontierScience, HLE w/o tools, SimpleQA, HealthBench Hard, HealthBench Professional, MedXpertQA (Text), FrontierScience Research, MMLU-Pro (Arcee), MMMLU
Scope: Frontier science, Broad academic knowledge, Factuality
Best Knowledge picks
BenchLM summaries for knowledge plus the practical tradeoffs users check next: open weights, price, speed, latency, and context.
Knowledge Leaderboard
Primary score: weighted knowledge score. Higher values rank first. Use the Show metric control to change the value shown in each row.
Switch between provisional-ranked and verified-ranked modes to compare the broader public dataset with sourced-only rankings.
Filters
1 Muse Spark 1.1 Meta | 93.5% | 74 | — | — | — | — | — | 62.1% | — | 52.2% | — | — | 59.3% | — | — | — | — |
2 Claude Opus 4.8 Anthropic | 92.5% | 82 | — | 93.6% | 93.6% | — | — | 57.9% | — | 49.8% | — | — | — | — | — | — | — |
3 Claude Sonnet 5 Anthropic | 91.2% | 79 | — | — | — | — | — | 57.4% | — | 43.2% | — | — | — | — | — | — | — |
4 Kimi K3 Moonshot AI | 89.5% | Est.77 | — | 93.5% | 93.5% | — | — | 56% | — | 43.5% | — | — | — | — | — | — | — |
5 Claude Opus 4.7 (Adaptive) Anthropic | 87.5% | 73 | — | 94.2% | 94.2% | — | — | 54.7% | — | 46.9% | — | — | — | — | — | — | — |
| 87% | 73 | — | 91.2% | 91.2% | — | — | 54.7% | — | 40.5% | — | — | — | — | — | — | — | |
7 Claude Opus 4.6 Anthropic | 85.8% | 69 | 99%P | 91.3% | 89.2% | 95% | 82% | 53% | 88%P | 40% | 72%P | 14.8% | — | 52.1% | — | 89.1% | — |
8 GPT-5.5 OpenAI | 83.5% | 74 | — | 93.6% | 93.6% | — | — | 52.2% | — | 41.4% | — | — | — | — | — | — | — |
9 GPT-5.6 Sol OpenAI | 83.2% | 80 | — | 94.6% | 94.6% | — | — | — | — | — | — | 33.1% | 60.5% | — | — | — | — |
10 GPT-5.4 OpenAI | 82% | 67 | 99%P | 92.8% | 92.8% | 96%P | 93%P | 52.1% | 91%P | 39.8% | 97%P | 40.1% | 48.1% | 59.6% | — | — | — |
11 GPT-5.6 Terra OpenAI | 81.5% | 77 | — | 92.9% | 92.9% | — | — | — | — | — | — | 32.7% | 57.7% | — | — | — | — |
12 GPT-5.2 OpenAI | 81% | 55 | 99%P | 92.4% | — | 95%P | 88%P | 42%P | 91%P | — | 95%P | — | — | — | — | — | — |
13 GPT-5.6 Luna OpenAI | 80.9% | 73 | — | 92.3% | 92.3% | — | — | — | — | — | — | 32.0% | 55.7% | — | — | — | — |
14 | 80.8% | 64 | 91.7%P | 86% | 86.0% | 66.8% | 85.7% | 50.4% | 74%P | — | 84%P | — | — | — | — | 85.8% | — |
15 Claude Sonnet 4.6 Anthropic | 79.7% | 64 | 99%P | 89.9% | — | 95% | 79.2% | 49% | 85%P | — | 48.5%P | — | — | — | — | — | — |
16 | 77.4% | 55 | — | 86.6% | — | 67.1% | 86.7% | — | — | — | — | — | — | — | — | — | — |
17 | 76.5% | 54 | — | 85.5% | — | 65.6% | 86.1% | — | — | — | — | — | — | — | — | — | — |
18 | 75.3% | 49 | — | 84.2% | — | 63.4% | 85.3% | — | — | — | — | — | — | — | — | — | — |
19 MiMo-V2.5-Pro Xiaomi | 74.8% | Est.60 | — | — | — | — | — | 48% | — | 34% | — | — | — | — | — | — | — |
20 Qwen3.7 Max Alibaba | 73.8% | 78 | — | 92.4% | 92.4% | 73.6% | 89.6% | 41.4% | — | — | — | — | — | — | — | — | 90.3% |
21 | 73.5% | 67 | — | 90.1% | 90.1% | — | 87.5% | 37.7% | — | — | 57.9% | — | — | — | — | — | — |
22 Gemini 3.1 Pro Google | 73% | 73 | — | 97%P | 94.3% | 95%P | 92%P | 40%P | 88%P | 45.4% | 95%P | 20.6% | — | 71.5% | — | — | — |
23 Claude Fable 5 Anthropic | 72.9% | 80 | — | 94.5%P | — | — | — | 64.5%P | — | 59%P | — | — | — | — | — | — | — |
| 72.5% | 62 | — | 87.9% | 87.9% | — | — | 46% | — | 30% | — | — | — | — | — | — | — | |
25 | 72.1% | Est.53 | 39%P | 77.5% | — | 62.6% | 83% | 1%P | 39%P | — | 54.3%P | — | — | — | — | — | — |
Top AI Models for Knowledge — July 2026
As of July 2026, Muse Spark 1.1 leads the provisional knowledge leaderboard with a score of 93.5%, followed by Claude Opus 4.8 (92.5%) and Claude Sonnet 5 (91.2%). BenchLM is currently showing 52 provisional-ranked models and 27 verified-ranked models in this category.
What changed
Claude Mythos Preview leads knowledge with the strongest HLE and FrontierScience scores.
GPT-5.4 close second, with excellent GPQA Diamond scores.
Claude Opus 4.6 holds #3, strong on SuperGPQA and SimpleQA factual accuracy.
Top models by benchmark
Expert-level questions in biology, physics, and chemistry(7% of category score)
Score in Context
What these scores mean
Knowledge carries a 12% weight in overall scoring. The weighted score blends expert-level tests (HLE, FrontierScience, GPQA) with broad knowledge (MMLU-Pro). A model scoring 90+ on MMLU might still struggle with research-level scientific reasoning — broad knowledge doesn't guarantee deep expertise.
Known limitations
MMLU is saturated — top models score 90%+, making it poor at differentiating. HLE ("Humanity's Last Exam") is deliberately very hard, so even top models score below 30%. Small score differences on HLE are noisy. SimpleQA measures factual accuracy but can penalize models that hedge appropriately.
How we weight
Knowledge carries a 12% weight in BenchLM.ai's overall scoring. A model scoring 90+ on MMLU might still struggle with research-level scientific reasoning — broad knowledge doesn't guarantee deep expertise.
For tasks like research assistance, factual Q&A, content creation, and educational applications, knowledge benchmark scores remain one of the strongest predictive signals. See the knowledge leaderboard for the top models in this category.
Leaderboards exclude benchmark rows that BenchLM generated from other scores or cloned from reference models. When a weighted benchmark is missing after that filter, the category falls back to the remaining trustworthy public rows instead of filling the gap with synthetic values.
The full scoring rules, freshness handling, and runtime/pricing caveats live on the BenchLM methodology page.
Scroll horizontally to read the full evidence ledger.
| Benchmark | Weight | Status | Description |
|---|---|---|---|
| MMLU | — | Display only | Tests knowledge across 57 academic subjects |
| GPQA | 7% | Weighted | Expert-level questions in biology, physics, and chemistry |
| GPQA-D | — | Display only | Provider-table reference for GPQA Diamond scores reported in first-party comparison charts. |
| SuperGPQA | 7% | Weighted | Enhanced version covering 285 disciplines |
| MMLU-Pro | 30% | Weighted | Harder version of MMLU with 10 answer choices and more reasoning-focused questions |
| HLE | 45% | Weighted | Extremely difficult questions contributed by domain experts worldwide to test frontier AI |
| FrontierScience | — | Display only | Research-level science and scientific reasoning benchmark |
| HLE w/o tools | — | Display only | Tool-free variant of Humanity's Last Exam used to isolate raw frontier reasoning without external aids |
| SimpleQA | 11% | Weighted | Factual question answering benchmark |
| HealthBench Hard | — | Display only | A harder health reasoning benchmark subset used in first-party frontier model comparisons. |
| HealthBench Professional | — | Display only | An open benchmark for clinician-facing model responses across care consult, writing and documentation, and medical research tasks. |
| MedXpertQA (Text) | — | Display only | Medical multiple-choice benchmark covering many specialties with text-only questions. |
| FrontierScience Research | — | Display only | A research-oriented FrontierScience variant focused on scientific investigation and solution quality. |
| MMLU-Pro (Arcee) | — | Display only | Display-only MMLU-Pro reference from Arcee AI's Trinity-Large-Thinking launch chart. |
| MMMLU | — | Display only | A multilingual MMLU-style benchmark reported in provider evaluation tables. |
About Knowledge Benchmarks
Tests knowledge across 57 academic subjects
Common questions
What is the best LLM for knowledge tasks?
The top LLMs for knowledge tasks are ranked by benchmarks like MMLU and GPQA, which test factual accuracy and expert-level understanding across dozens of subjects.
What is MMLU and how does it measure LLM knowledge?
MMLU (Massive Multitask Language Understanding) tests LLMs across 57 subjects from STEM to humanities, measuring broad factual knowledge and reasoning at varying difficulty levels.
What benchmarks test knowledge in AI models?
Key knowledge benchmarks include MMLU, MMLU-Pro, GPQA, SuperGPQA, HLE, and FrontierScience, each evaluating different depths of factual and scientific understanding.
How do knowledge benchmarks differ from reasoning benchmarks?
Knowledge benchmarks focus on factual recall and domain expertise, while reasoning benchmarks test logical deduction and multi-step problem solving independent of specific facts.
Knowledge benchmark updates
Know which model knows the most — updated every week.
One email each week. Unsubscribe anytime.