The Frontier · Intelligence · Metric

MRCR 1M mean match ratio

mrcr_1m_mmr · unit mmr_score · higher is better · Model capabilities

Evaluation contexts differ across rows — do not treat values as a universal ranking.

EntityValueUnitAs ofOriginVerificationContextSourceEvidenceArticle
Claude Opus 4.692.9mmr_score—paperReported (unverified)MRCR 1M — Claude Opus 4.6 Max (same Table 6)arXiv — DeepSeek-V4 (2606.19348)Table 6 — MRCR 1M MMRmillion-token-mrcr-claim-check
DeepSeek-V4-Pro83.5mmr_score—paperReported (unverified)MRCR 1M — Think Max / Max modearXiv — DeepSeek-V4 (2606.19348)Table 6 — MRCR 1M MMRmillion-token-mrcr-claim-check
DeepSeek-V4-Pro44.7mmr_score—paperReported (unverified)MRCR 1M — Non-Think modearXiv — DeepSeek-V4 (2606.19348)Table 7 — Non-Think modemillion-token-mrcr-claim-check