LLM: Large Language Model
LLM (Large Language Model) — a transformer neural network with tens of billions to trillions of parameters, trained on a massive text corpus. Generates human-like output for any NLP task: chat, code, summarise, translate. 2026 leaders: GPT-5 / GPT-5 Pro (OpenAI), Claude Opus 4.7 / Sonnet 4.6 (Anthropic), Gemini 2.5 (Google), Llama 3 (Meta open), DeepSeek R1 (open). Parameters: 8B → 1.8T. Context window: 200k → 2M tokens.
Below: details, example, related terms, FAQ.
Details
- Architecture: Transformer (Vaswani et al 2017), decoder-only in most models
- Training: pretrain on text corpus (TB) + RLHF + instruction tuning
- Parameters (2026): small 8B, mid 70B, frontier 400B+ (sparse MoE equivalent of 1.8T)
- Inference cost: $0.15-15 per 1M tokens (API) vs self-host (A100/H100/B100)
- Context window 2026: Claude 1M stable, Gemini 2M, GPT-5 also 1M
Example
# OpenAI API call
curl https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
"model": "gpt-5",
"messages": [{"role": "user", "content": "Explain TCP vs UDP"}]
}'Related Terms
Learn more
Frequently Asked Questions
Open-source LLM?
Llama 3 (Meta), Mistral Large, Qwen 2.5, DeepSeek R1 — free weights, MIT/Apache. Performance is approaching GPT-5.
Self-host cost?
70B model needs a server with 2× H100 (~$80k) or cloud GPU $5/h. ROI at > 10M tokens/day.
Hallucinations?
LLMs generate likely text, not facts. Mitigations: RAG (grounding) + fact-check output + low temperature (0.1-0.3 for facts).