OpenAI confirmed in a Wednesday post on its developer blog that GPT-5 will begin a phased enterprise rollout in Q3 2026 with broader consumer access opening in Q4. The model has been in red-team and trusted-tester evaluation since late February. Internal benchmarks shared with selected enterprise customers show step-function gains over GPT-4.5 on long-context reasoning, coding, and agentic task completion, with smaller but meaningful gains on standard math and language benchmarks.
The capability profile is where this matters. GPT-5 reportedly handles 2 million token context windows natively, which is the same range Google announced for Gemini 3.1 Ultra last week. The agentic task completion score on the OpenAI internal benchmark went from 64 percent on GPT-4.5 to 87 percent on GPT-5, which crosses the threshold most enterprise customers have specified for production deployment of agentic workflows. The coding score on SWE-bench Verified went from 44 percent to 71 percent.
Pricing for the API will start at 12 dollars per million input tokens and 60 dollars per million output tokens for the base model and 80 dollars per million input tokens and 240 dollars per million output tokens for the high-reasoning variant called GPT-5 Pro. Those numbers undercut the Anthropic Claude Opus 4.6 pricing of 80 dollars per 250 dollars per million tokens for similar capability tiers. The Microsoft Copilot integration is expected to land within 30 days of general availability.
The enterprise deployment plan that has surfaced in customer briefings is structured. JPMorgan Chase, Bristol Myers Squibb, McKinsey, Bain, and a small number of named law firms are in the first wave with August deployment targets. The second wave includes broader Fortune 500 access starting in September. The Microsoft 365 Copilot integration for the broader enterprise market is targeted for October. Consumer ChatGPT access for Plus and Pro subscribers begins in November.
The competitive context is the most interesting part. Anthropic released Claude Opus 4.6 to enterprise customers in late April with similar pricing and capability claims on certain benchmarks. Google released Gemini 3.1 Ultra Tuesday morning with native multimodality and a 2 million token window at lower API pricing of 12 and 60 dollars per million tokens. DeepSeek launched V4 Flash and V4 Pro last month at a meaningful discount to all three U.S. labs. The market is now four major frontier model families competing on capability and price.
What is real for businesses planning AI deployments. The total spend on frontier model capacity by enterprises in 2026 will likely cross 60 billion dollars, up from 28 billion in 2025. The marginal cost per query has fallen 47 percent year over year for equivalent capability. The gap between the top three labs has narrowed on most public benchmarks but remains wide on specific verticals including legal reasoning, complex coding, and long-context document analysis.
For small businesses, the practical question is which subscription stack to run. The current Lumina Media production stack of three tools at 187 dollars per month including ChatGPT Pro at 200, Claude Pro at 30, and Copilot at 30 produces most of the value most teams need. The September enterprise rollout will likely push subscription costs up roughly 18 to 24 percent for the top tiers. ChatGPT Plus is expected to remain at 25 dollars per month with GPT-5 base access. The new GPT-5 Pro tier in ChatGPT will be 60 dollars per month.
The capability that will move enterprise budgets is agentic workflows. GPT-5 reportedly completes 14 percent of the multi-step browser tasks on OpenAI's WebArena benchmark in fewer than three retries, up from 4 percent on GPT-4.5. The number sounds small but it crosses the threshold for production use. The first wave of agent deployments at scale will be in Q4 2026 across customer service, document review, and routine data analysis. Anthropic has held a lead on computer use agents through 2025 but the GPT-5 results suggest the gap closes significantly.
The Apple integration question remains open. Apple Intelligence V2 launches at WWDC June 8 with on-device models and reportedly an upgraded ChatGPT integration. Apple has not confirmed whether the WWDC integration will use GPT-4.5 or wait for GPT-5. The early Apple Intelligence integration ran on GPT-4o for the first six months before transitioning to GPT-4.5 in February.
For Wesley Insider readers running content businesses, the practical timeline is clear. Most of 2026 will run on GPT-4.5 and Claude Opus 4.6 capability. The next wave of capability becomes generally available in Q4. The differentiator will not be which model anyone uses but how well the workflow uses it. A small business with 24 to 48 hours of monthly model usage on a thoughtful workflow stack will produce more value than a competitor with 200 hours of unstructured usage on the same models.
The Q1 cloud reports last week confirm the broader spending pattern. Microsoft Azure grew 37 percent. Google Cloud grew 32. AWS grew 19. The bulk of that growth is attributable to AI workloads. The 700 billion dollars in capital expenditure across the four largest U.S. tech firms is being deployed primarily into Nvidia silicon and the power infrastructure required to operate it.