MediumReasoningReviewer-confirmedPublished

Lost in the middle: degraded recall for mid-context information

Retrieval accuracy is highest for facts at the start and end of a long context and drops for facts in the middle.

Published June 26, 2026

Reproducibility: Often
Severity: Medium
Confidence: Reviewer-confirmed

Details

Performance on long-context retrieval follows a U-shaped curve: information placed in the middle of a long input is recalled far less reliably than information at the beginning or end. This directly impacts RAG and long-document QA.

Found with

🔬 Needle-in-a-haystack (long-context retrieval)

Vary needle depth; accuracy dips mid-context.

🔬 Boundary & edge-case testing

Evidence

https://arxiv.org/abs/2307.03172

Liu et al., 'Lost in the Middle' (2023)

Affected versions

Anthropic · claude-opus-4-8OpenAI · gpt-4oGoogle · gemini-2.0-flashMeta · llama-3.3-70b

References

Lost in the Middle: How Language Models Use Long Contexts

RAG Context window

Source: https://arxiv.org/abs/2307.03172

Cite this

Qlarify Labs. (2026). Lost in the middle: degraded recall for mid-context information. Retrieved from https://labs.qlarify.fi/findings/lost-in-the-middle