OpenAI's GPT-5 is now reaching enterprise customers through a staged rollout that began in mid-July 2026, giving large organisations access to substantially improved reasoning, longer context windows, and native agentic capabilities that were absent from its predecessor.
The release extends the effective context window to 256,000 tokens in standard API access, with experimental 1M-token access available to select research partners. OpenAI says the expanded context eliminates the need for most retrieval-augmented generation architectures in internal document workflows, because entire contract repositories or codebases can fit within a single prompt.
The most operationally significant addition is the redesigned tool-calling interface, now branded as "Agent Chains." Unlike the single-round function call pattern in GPT-4o, Agent Chains allow the model to execute multi-step sequences across registered tools — searching a database, summarising findings, drafting an email, and scheduling a calendar event — in a single user request, with human-in-the-loop checkpoints configurable at the API level.
Early enterprise adopters in legal, healthcare, and financial services report cut in manual review time by 40–60% for document-heavy workflows. A large US law firm trialling the system noted that contract review tasks that previously required a paralegal full day could be completed in under two hours with appropriate human spot-checks.
Pricing for the enterprise tier reflects the increased capability. GPT-5 Turbo (the default enterprise model) is priced at $10 per million input tokens and $30 per million output tokens — roughly 3x the cost of GPT-4o at equivalent volume tiers. OpenAI is offering migration credits to enterprise customers who transfer workflows from GPT-4 to GPT-5 before September 2026.
Security posture has also been hardened. GPT-5 enterprise API calls can be scoped to organisation-specific "knowledge domains" that prevent the model from reasoning over topics outside a configured subject area, a feature aimed at regulated industries where scope creep in AI responses creates compliance risk.
OpenAI CEO Sam Altman described GPT-5 as the company's first "truly agentic-first model," noting that the architecture was designed from the ground up to handle multi-turn tool use rather than retrofitting it onto a language-first design.
The broader developer API access is expected to follow in August 2026, with GPT-5 Mini — a cost-optimised derivative — coming in the same window for high-volume, low-latency use cases.