GPT-5.6 Sol preview: capabilities, safeguards, and availability

Overview

OpenAI announced a limited preview of GPT‑5.6 Sol, the flagship model of the GPT‑5.6 series, together with Terra and Luna models, to evaluate performance and safety before broader release.

Capabilities

GPT‑5.6 Sol delivers the strongest performance yet in coding, biology, and cybersecurity workflows. It sets a new state of the art on Terminal‑Bench 2.1, improves results on GeneBench v1 using fewer tokens, and shows competitive results on ExploitBench and ExploitGym while using less output than prior models. The model introduces a new max reasoning effort for deeper single‑agent reasoning and an ultra mode that uses subagents to accelerate complex work.

Safeguards

The release pairs increased capabilities with a layered safeguard stack designed to curb prohibited offensive cyber activity while preserving legitimate defensive work. Safeguards include model‑level refusal training, real‑time misuse classifiers, account‑level review, differentiated access, monitoring, and enforcement, all pressure‑tested via automated red‑teaming (over 700,000 A100‑equivalent GPU hours) and human expert testing. GPT‑5.6 Sol does not cross the Cyber Critical threshold under the Preparedness Framework. During the preview, safeguards may block or delay requests, and feedback will be used to reduce unnecessary interference.

Availability and Pricing

Initially, GPT‑5.6 Sol, Terra, and Luna are available via the API and Codex to a select group of trusted partners; broader availability through ChatGPT, Codex, and the API is planned for the coming weeks. Pricing per 1M tokens is Sol: $5 input / $30 output, Terra: $2.50 input / $15 output, Luna: $1 input / $6 output. Terra has competitive performance to GPT‑5.5 while being 2x cheaper and Luna brings strong capability at the lowest cost. The release also introduces more predictable prompt caching with explicit breakpoints and a 30‑minute minimum cache life, where cache writes cost 1.25× the uncached input rate and cache reads receive a 90% discount. GPT‑5.6 Sol will run on Cerebras hardware at up to 750 tokens per second starting in July, initially for select customers.