Chinchilla
by Google DeepMind · paid · Last verified 2026-03-17
Chinchilla is DeepMind's 70 billion parameter language model that introduced the landmark 'Chinchilla scaling laws' showing that most large language models are significantly undertrained relative to their compute budget. By training on 1.4 trillion tokens with only 70B parameters, Chinchilla outperformed the 280B Gopher model while using the same compute, reshaping how the entire field approaches LLM training.
https://arxiv.org/abs/2203.15556 ↗C+
C+—Average
Adoption: CQuality: AFreshness: DCitations: A+Engagement: F
Specifications
- License
- Proprietary
- Pricing
- paid
- Capabilities
- text-generation, question-answering, language-understanding, few-shot-learning
- Integrations
- Use Cases
- LLM scaling research, compute-optimal training studies, NLP benchmark evaluation, AI policy research
- API Available
- No
- Parameters
- ~70B
- Context Window
- 2K
- Modalities
- text
- Training Cutoff
- 2022
- Tags
- foundational, deepmind, scaling-laws, compute-optimal, historical
- Added
- 2026-03-17
- Completeness
- 100%
Index Score
56Adoption
40
Quality
85
Freshness
30
Citations
92
Engagement
0