Zero-shot
Zero-shot means asking a model to do a task from an instruction alone, with no worked examples in the prompt and no task-specific training. The model relies only on what it already learned.
Also called: zero-shot learning, zero-shot prompting
By Divya Sharma · Checked Nov 15, 2026 · How we review
A zero-shot prompt describes the task and gives the input, nothing more. "Translate this sentence into German" or "Is this review positive or negative?" are zero-shot prompts. Most everyday use of a chat assistant is zero-shot.
Zero-shot, one-shot and few-shot
The terms count examples. Zero-shot uses none, one-shot uses one and few-shot prompting uses several. Adding examples usually helps when the output format is unusual or the task is ambiguous, at the cost of more input tokens on every call.
Zero-shot reasoning
A 2022 paper by Kojima and colleagues showed that adding one line, "Let's think step by step", made large models better at reasoning tasks without any hand-written examples. They called this Zero-shot-CoT, a zero-shot form of chain-of-thought prompting.
A different meaning on leaderboards
Outside prompting, zero-shot also describes evaluation: how a model performs on tasks or datasets it was not trained on. Embedding leaderboards such as MTEB use the word in this sense, so a zero-shot figure there is a statement about overlap between training data and test data, not about prompts. A model that saw a benchmark's training data may score well on it without generalising as well. See benchmark contamination.
Why it matters
Starting zero-shot is the cheapest baseline. If it works, you avoid maintaining examples. If it fails, the failure tells you what examples or further prompt engineering need to fix.
Related terms
- Few-shot prompting
- Few-shot prompting is giving a language model a handful of worked examples of a task inside the prompt, so it can copy the pattern for a new input without any retraining.
- Chain-of-thought
- Chain-of-thought (CoT) is a prompting method in which a language model writes out intermediate reasoning steps before its final answer, which tends to improve results on maths, logic and other multi-step problems.
- Prompt engineering
- Prompt engineering is the practice of writing, structuring and testing the instructions and examples given to a language model so that it produces the output an application needs, without changing the model's weights.
- MTEB
- MTEB, the Massive Text Embedding Benchmark, is a public benchmark and leaderboard that scores text embedding models across many task types, such as retrieval, clustering, classification and reranking, over many datasets and languages.
- Benchmark contamination
- Benchmark contamination is when a model has seen a benchmark's test questions or answers during training, so its score on that benchmark overstates how well it handles new problems of the same kind.