Sapience scores 97.7 on MRCR reading only ~15K tokens, beating GPT-6 Astra's 96.3
gordic_aleksa · x · 2026-09-15
Aleksa Gordić shared a surprising long-context result: on OpenAI's MRCR benchmark, Sapience Labs' custom system scores 97.7 versus 96.3 for GPT-6 Astra.
The key difference is the approach: Sapience reads only 15K tokens and lets the open model DeepSeek V4 Pro do the answering, while Astra has to ingest the full 500K-1M token context. Gordić says he's unaffiliated—the system is built by @lvrzhn—but finds the result promising.
Related event: Sapience tops GPT-6 Astra on MRCR reading just 15K tokens(2 posts)→
More from Models
- Perplexity says GPT-6 Astra can run full workflows with far fewer human check-ins — imjustnewatai · 2026-09-15
- Tracing 'seam': how Claude's vocabulary drift spread across 685K GitHub PRs — dl_weekly · 2026-09-15
- OpenAI flags compiler debugging with lldb as 'cybersecurity', sparking criticism — QuixiAI · 2026-09-15
- Kimi's AI 'escape containment' was a misconfigured sandbox, not an escape — ns123abc · 2026-09-15
- Nahcrof exposed for routing K2.5 to GPT OSS 20B in alleged token fraud — realmrfakename · 2026-09-15
- AI-text detector Pangram is powerful but badly calibrated, dev argues — maksym_andr · 2026-09-15