Large Language Model (LLM)
A Large Language Model (LLM) is a deep neural network — typically a transformer with billions to trillions of parameters — trained on large text corpora to predict the next token, then fine-tuned for instruction-following, dialogue, and increasingly tool use and reasoning.
The frontier LLMs as of April 2026 are Anthropic Claude Opus 4.7, OpenAI GPT-5.5, and Google Gemini 3.1 Pro. All three are decoder-only transformers; all three offer tool use, long context (200K to 1M+ tokens), and image input; none publish parameter counts, but inference behavior suggests trillions for the top tier.
Open-weights leaders include Meta Llama 4, Google Gemma 4 (April 2026), Qwen 3.5, and Mistral Large 2. Open-weights models lag the frontier on raw intelligence by roughly 6-12 months but lead on deployment flexibility (single-GPU inference, edge deployment, on-premise compliance).
Practical model selection depends on three axes: intelligence (frontier vs open-weights), cost-per-task (varies 10-100×), and deployment shape (cloud API, self-hosted, on-device). Koenig AI Academy publishes an updated frontier-model comparison at /data/claude-tool-use-determinism/.
## Related Terms
- [[glossary/transformer|Transformer]] — the neural architecture underlying virtually all modern LLMs - [[glossary/tokenization|Tokenization]] — related concept that intersects with this term in agent workflows - [[glossary/embedding|Embedding]] — dense vector representations that power semantic similarity search - [[glossary/fine-tuning|Fine-tuning]] — the weight-update process that adapts a pre-trained base model for downstream tasks - [[courses/claude-tool-use-from-zero|Course: Claude Tool Use from Zero]] — hands-on practice with the concepts covered in this entry