GPT-5.6 Sol (OpenAI)

According to the TechTimes report, OpenAI and Cerebras Systems announced an "Ultrafast" API service tier on August 13, 2026 that runs GPT-5.6 Sol on Cerebras wafer-scale silicon at a vendor-claimed 750 output tokens per second, which OpenAI describes as up to 14x faster than its Standard processing tier. The piece conf Ultrafast is positioned as a serving tier, not a new model: GPT-5.6 Sol's intelligence, context window, and output quality are stated to be unchanged, with only token delivery speed affected. The preview is invite-only with no published pricing or general-availability date, and OpenAI has not disclosed any Ultrafast pr

Per the BenchLM profile, GPT-5.6 Sol reached general availability on July 9, 2026 as a proprietary reasoning model with a 1.05M-token context window and a reported API price of $4 per million input tokens and $20 per million output tokens, with cached input at $0.40 and a blended cost near $12. The page assigns it a ca The profile also notes that GPT-5.6 Sol has since been superseded by OpenAI's newer GPT-6 Astra line, which reduces its current news value for fresh coverage. Its reported percentile profile stands at 97th in Agentic, 97th in Coding, 42nd in Reasoning, 98th in Knowledge, and 94th in Multimodal, with Math, Multilingual,

OpenAI began a limited preview of GPT-5.6 on June 26, 2026, packaged as a family of three durable tiers rather than a single model: Sol (flagship for ambitious agentic work), Terra (balanced everyday tier at roughly half of GPT-5.5's cost), and Luna (fast, low-cost option for high-volume jobs). In OpenAI's framing, the The preview guide reports Sol pricing at $5/$30 per million tokens — the same as GPT-5.5, framed as a free upgrade for that tier — alongside a preview-era Terminal-Bench 2.1 score of 91.9 in Sol's ultra mode, described as an OpenAI-reported SOTA figure rather than independently audited. Safety review consumed 700K+ A10

The BenchmarkList page dedicated to GPT-5.6 Sol tracks 196 benchmark rows with explicit GPT-5.6 Sol entries across agentic, coding, reasoning, and search evaluations. Sol holds field-leader status on multiple evals: BrowseComp at 92.2% (with X-high effort at 90.6% ±4.4% and Max effort at 90.4%), WildClawBench overall a Per-category aggregates show GPT-5.6 Sol as an agentic-and-coding leader with comparatively weaker reasoning: Agentic rank 6 of 154 (97th percentile, 9 benchmarks), Coding rank 5 of 156 (97th, 12 benchmarks), and Reasoning rank 11 of 19 (44th, 3 benchmarks). Reported row publication dates span 2026-07-09 through 2026-0