

Sol Ultrafast
#12 en Modèles de reasoningUnknown · depuis 13. August 2026 (limitierte Preview) · 2× · vu le 21 août 2026
Sol Ultrafast (officially "GPT-5.6 Sol Ultrafast" or "GPT-5.6 Sol on Ultrafast mode") is a new OpenAI API service tier for the reasoning model GPT-5.6 Sol, powered by Cerebras wafer-scale hardware. It delivers the same model intelligence as the standard tier but at up to 14x the speed and up to 750 output tokens per second. The service was announced on August 13, 2026 as a limited preview for select OpenAI API customers; official pricing and a general-availability date have not yet been published. In a Cerebras-run benchmark, Sol Ultrafast answered all 2,500 Humanity's Last Exam questions in 11 hours 11 minutes with accuracy comparable to Claude Fable 5.
Fonctionnalités
| Key Benchmark (%) | Humanity's Last Exam: solved all 2,500 questions in 11h 11m, ~7x faster than Claude Fable 5 at comparable accuracy; GDP-Val: 5.6x speedup with no quality loss |
| Context Window (Tokens) | 1,050,000 tokens (base GPT-5.6 Sol), 128,000 tokens max output |
| License | Proprietary, closed model, access only via OpenAI API preview for select customers |
| Multimodality | Text and image input, text output (inherited from GPT-5.6 Sol) |
| Platform | OpenAI API (limited preview, powered by Cerebras Wafer-Scale Engine) |
| Price per 1M Tokens | Not yet published (preview with no rate card); standard GPT-5.6 Sol costs $5.00 input / $30.00 output per 1M tokens |
| Release Date | August 13, 2026 (limited preview) |