reasoning benchmark
HellaSwag Leaderboard
A challenging commonsense natural language inference dataset that uses Adversarial Filtering to create questions trivial for humans (>95% accuracy) but difficult for state-of-the-art models, requiring completion of sentence endings based on physical situations and everyday activities
Updated Aug 17, 2026
Models27
Model coverage27
MetricScore
EvidenceB
HellaSwag Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelAN | Score95.4% | Percentile100.0% | Participants27 | EvidenceC | Evaluated |
| Rank02 | ModelOP | Score95.3% | Percentile96.2% | Participants27 | EvidenceC | Evaluated |
| Rank03 | ModelGO | Score93.3% | Percentile92.3% | Participants27 | EvidenceC | Evaluated |
| Rank04 | ModelXI | Score89.8% | Percentile88.5% | Participants27 | EvidenceC | Evaluated |
| Rank05 | ModelAN | Score89.0% | Percentile84.6% | Participants27 | EvidenceC | Evaluated |
| Rank06 | ModelCO | Score88.6% | Percentile80.8% | Participants27 | EvidenceC | Evaluated |
| Rank07 | ModelNR | Score88.2% | Percentile76.9% | Participants27 | EvidenceC | Evaluated |
| Rank08 | ModelAC | Score87.6% | Percentile73.1% | Participants27 | EvidenceC | Evaluated |
| Rank09 | ModelGO | Score86.5% | Percentile69.2% | Participants27 | EvidenceC | Evaluated |
| Rank10 | ModelGO | Score86.4% | Percentile65.4% | Participants27 | EvidenceC | Evaluated |
| Rank11 | ModelAN | Score85.9% | Percentile61.5% | Participants27 | EvidenceC | Evaluated |
| Rank12 | ModelNV | Score85.6% | Percentile57.7% | Participants27 | EvidenceC | Evaluated |
| Rank13 | ModelAC | Score85.2% | Percentile53.9% | Participants27 | EvidenceC | Evaluated |
| Rank14 | ModelMI | Score83.8% | Percentile50.0% | Participants27 | EvidenceC | Evaluated |
| Rank15 | ModelMA | Score83.5% | Percentile46.1% | Participants27 | EvidenceC | Evaluated |
| Rank16 | ModelAC | Score83.0% | Percentile42.3% | Participants27 | EvidenceC | Evaluated |
| Rank17 | ModelGO | Score81.9% | Percentile38.5% | Participants27 | EvidenceC | Evaluated |
| Rank18 | ModelIB | Score80.1% | Percentile34.6% | Participants27 | EvidenceC | Evaluated |
| Rank19 | ModelGO | Score78.6% | Percentile30.8% | Participants27 | EvidenceC | Evaluated |
| Rank20 | ModelGO | Score78.6% | Percentile26.9% | Participants27 | EvidenceC | Evaluated |
| Rank21 | ModelAC | Score76.8% | Percentile23.1% | Participants27 | EvidenceC | Evaluated |
| Rank22 | ModelGO | Score72.2% | Percentile19.2% | Participants27 | EvidenceC | Evaluated |
| Rank23 | ModelGO | Score72.2% | Percentile15.4% | Participants27 | EvidenceC | Evaluated |
| Rank24 | ModelME | Score69.8% | Percentile11.5% | Participants27 | EvidenceC | Evaluated |
| Rank25 | ModelMI | Score69.4% | Percentile7.7% | Participants27 | EvidenceC | Evaluated |
| Rank26 | ModelMI | Score69.1% | Percentile3.9% | Participants27 | EvidenceC | Evaluated |
| Rank27 | ModelBA | Score33.0% | Percentile0.0% | Participants27 | EvidenceC | Evaluated |
HellaSwag Highlights
The leading models and scores on this benchmark.
HellaSwag Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for HellaSwag
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
Ranking basisThis hellaswag AI model leaderboard uses descending score in the benchmark's original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.
- 01ANClaude 3 OpusAnthropic
- Score
- 95.4%
- Speed
- Up to 120 tok/s via Bedrock
Strengths
- Ranks #1 of 27 compared models
- 100th percentile on this benchmark
- C evidence result
Considerations
- This result measures HellaSwag, not total model capability
- 02OPGPT-4OpenAI
- Score
- 95.3%
- Price
- $30 input / $60 output per 1M tokens
- Speed
- Up to 104 tok/s via Azure
Strengths
- Ranks #2 of 27 compared models
- 96th percentile on this benchmark
- C evidence result
Considerations
- This result measures HellaSwag, not total model capability
- 03GOGemini 1.5 ProGoogle
- Score
- 93.3%
- Speed
- Up to 85 tok/s via Google
Strengths
- Ranks #3 of 27 compared models
- 92th percentile on this benchmark
- C evidence result
Considerations
- This result measures HellaSwag, not total model capability
- 04XIMiMo-V2.5-ProXiaomi
- Score
- 89.8%
- Price
- $0.43 input / $0.87 output per 1M tokens
- Speed
- Up to 69 tok/s via Novita
Strengths
- Ranks #4 of 27 compared models
- 88th percentile on this benchmark
- C evidence result
Considerations
- This result measures HellaSwag, not total model capability
- 05ANClaude 3 SonnetAnthropic
- Score
- 89.0%
- Speed
- Up to 120 tok/s via Bedrock
Strengths
- Ranks #5 of 27 compared models
- 85th percentile on this benchmark
- C evidence result
Considerations
- This result measures HellaSwag, not total model capability
Selection summary
Best AI Models for HellaSwag
Claude 3 Opus currently leads HellaSwag with 95.4%. It is the top model on this specific benchmark, while the best LLM for the broader task should also be checked against other benchmarks, price and runtime.
Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.
What is HellaSwag?
What HellaSwag measures and how its scores work.
A challenging commonsense natural language inference dataset that uses Adversarial Filtering to create questions trivial for humans (>95% accuracy) but difficult for state-of-the-art models, requiring completion of sentence endings based on physical situations and everyday activities
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- HellaSwag
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- hellaswag|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about HellaSwag.
Which model scores highest on HellaSwag?
Claude 3 Opus is currently ranked first with 95.4%.
What does HellaSwag measure?
A challenging commonsense natural language inference dataset that uses Adversarial Filtering to create questions trivial for humans (>95% accuracy) but difficult for state-of-the-art models, requiring completion of sentence endings based on physical situations and everyday activities
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
27 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.