Needle-in-a-Haystack (long-context retrieval probe) — Anthropic explicit 8-needle, 1M-context config; distinct from unscoped entry.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.6 | Anthropic | 76 | 1 | 2026-06-06 |
| 2 | Claude Sonnet 4.5 | Anthropic | 18.5 | 1 | 2026-06-06 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.