Sapience reportedly scores 97.7 on MRCR versus GPT-6 Astra’s 96.3
A user says Sapience reads about 15,000 tokens and uses the open model DeepSeek V4 Pro to answer, while GPT-6 Astra reads 500,000–1 million tokens.
TLDR
A user sharing long-context benchmark results says Sapience scores 97.7 on OpenAI’s MRCR, compared with 96.3 for GPT-6 Astra. They describe Sapience as a custom system that reads about 15,000 tokens, with DeepSeek V4 Pro doing the answering; Astra, they say, reads the full 500,000–1 million tokens. In a follow-up reply, the same user cites similar results on RULER and suggests Sapience does not appear custom-built to game a particular long-context benchmark.
Combined views
—
2 Sources, first seen ago
