OpenAI GPT-5.6 Sol Ultrafast: 14x Speed Boost with Cerebras Wafer-Scale Chips

Technology: Cerebras Wafer-Scale Chips

Ultrafast is powered by Cerebras wafer-scale chips (Cerebras CS-3 systems) . Cerebras and OpenAI co-announced the partnership, with Cerebras providing the low-latency inference hardware that makes the speedup possible . This is not a new model—Ultrafast runs the existing GPT-5.6 Sol, previewed in June 2026 and broadly available since July, on specialized Cerebras infrastructure .

Target Corporate Workflows

OpenAI is targeting enterprise workflows where every millisecond matters . During the preview, customers are testing Ultrafast across five primary verticals :

• Incident response — Real-time triage and remediation commands, analyzing application logs, recent code changes, and engineer reports to identify root causes during outages
• Customer support — Instant, high-volume conversational responses
• Financial analysis — Rapid market data interpretation, transaction assessment, and report generation
• Coding and software development — Near-instant code generation, review, and debugging
• Commerce — Real-time product recommendations and dynamic pricing

OpenAI is also applying Ultrafast to research workflows involving knowledge searches, data queries, connected tools, and information synthesis .

Current Availability and Pricing

Ultrafast is in limited preview (waitlist-based) for select API customers as of August 13–14, 2026 . OpenAI says access will expand to more businesses as Cerebras capacity grows . Pricing has not been disclosed—only that it will be a premium API tier . Standard GPT-5.6 API prices remain Sol $5/$30, Terra $2/$12, and Luna $0.20/$1.20 per million input/output tokens .

The limited preview reflects OpenAI's cautious approach to scaling a new infrastructure partnership. As capacity on Cerebras systems grows, wider access is expected, but no timeline has been announced .