Best Models for Long-Context Retrieval - March 2026
Claude Opus 4.6 leads multi-needle retrieval at 1M tokens with 76% on MRCR v2, while GPT-5.4 achieves near-perfect single-needle accuracy across its full 1M context.
Mrcr
Claude Opus 4.6 leads multi-needle retrieval at 1M tokens with 76% on MRCR v2, while GPT-5.4 achieves near-perfect single-needle accuracy across its full 1M context.
