Cerebras Powers Ultrafast Mode for OpenAI’s GPT-5.6 Sol

Published: August 13, 2026 — Cerebras (NASDAQ: CBRS) / GlobeNewswire

Official vendor announcement:

• Ultrafast service tier in OpenAI API for GPT-5.6 Sol
• Up to 750 output tokens/sec; up to 14× Standard processing
• Limited preview for select OpenAI customers; expand with capacity
• Same intelligence as Standard (vendor claim)
• Artificial Analysis–based comparisons: 5× Claude Opus 4.8 Fast; 11× Claude Fable 5
• HLE full set: ~11 hours Ultrafast vs >3 days Claude Fable 5 (~7×)
• GDP-Val: 5.6× end-to-end speedup, no quality loss (vendor claim)
• Wafer-Scale Engine: 44 GB on-chip SRAM; weights stay on-chip

Quotes: Andrew Feldman (Cerebras CEO); Sachin Katti (OpenAI VP Compute Strategy & GPT-Infra)

Speed/quality multiples are vendor + Artificial Analysis cited — not independent peer review.