Skip to content
News Model release

GPT-5.5 reaches general availability in Microsoft Foundry at $5 per million input tokens

· by Pondero Newsdesk

The short version

OpenAI's GPT-5.5 went generally available in Microsoft Foundry on April 24, 2026, with a Pro tier priced at $30 per million input tokens for the most demanding enterprise agent workloads.

GPT-5.5 reaches general availability in Microsoft Foundry at $5 per million input tokens

OpenAI's GPT-5.5 became generally available in Microsoft Foundry on April 24, 2026, bringing the model's agentic capabilities to enterprise teams building on Azure, per the Azure blog announcement published April 23.

What

The release ships two tiers. Per Microsoft's pricing table, GPT-5.5 is priced at $5.00 per million input tokens, $0.50 per million cached input tokens, and $30.00 per million output tokens. GPT-5.5 Pro, targeted at higher-complexity workloads, costs $30.00 per million input tokens, $3.00 per million cached input tokens, and $180.00 per million output tokens. Both are eligible for Microsoft Azure Consumption Commitment (MACC) spending, per the same Azure blog post.

Per the Azure blog, GPT-5.5 advances the GPT-5 arc in four areas: deeper long-context reasoning, more reliable agentic execution, improved computer-use accuracy, and greater token efficiency. GPT-5.4 had introduced multi-step reasoning and early agentic capabilities. GPT-5.5 extends that for "sustained, high-stakes professional workflows," according to the same Microsoft post.

The model integrates with Foundry Agent Service, which provides isolated sandbox environments for each deployed agent, including a persistent filesystem, a dedicated Microsoft Entra identity, and scale-to-zero pricing. Declarative agents defined in YAML or built with frameworks including LangGraph, Claude Agent SDK, and OpenAI Agents SDK can all run on the service, per the Azure announcement. Microsoft highlights several target domains: software engineering, DevOps, legal, health sciences, and professional services. Those are the areas where the cost of imprecision is high and where sustained, multi-turn execution matters more than raw single-query speed.

Naomi Moneypenny, Senior Director Product Development at Microsoft, authored the Azure blog post. The post frames the release as "frontier intelligence on an enterprise ready platform," positioning it squarely at large-org buyers rather than individual developers or startups.

Why it matters

For enterprise teams running agent pipelines in production, the pricing structure tells most of the story. GPT-5.5 at $5/$30 per million tokens sits meaningfully above GPT-4o's standard tiers but below the Pro variant, per the Azure pricing table. Teams choosing the Pro tier at $30/$180 are targeting workloads where output quality justifies the cost, such as legal research, software engineering at repo scale, or document synthesis across large corpora.

The computer-use improvements are a practical concern for anyone running agents against internal web portals, dashboards, or admin interfaces that lack clean API surfaces. Per the Azure blog, Microsoft describes GPT-5.5 as capable of "navigating software interfaces with improved precision and more reliable recovery when execution takes an unexpected turn." That framing is a vendor claim, not a third-party benchmark result.

Per the same Microsoft pricing disclosure, MACC eligibility means enterprise customers with existing Azure spending commitments can fold GPT-5.5 usage into existing contract draw-downs rather than separate billing. That matters for procurement teams managing multi-year cloud agreements. The token caching pricing, $0.50 per million for GPT-5.5 and $3.00 per million for GPT-5.5 Pro, is specifically relevant for agentic workflows that repeatedly send large system prompts or shared context across many parallel agent runs.

Context and reactions

GPT-5.5 is the third model in a visible progression. GPT-5 unified reasoning and speed. GPT-5.4 added multi-step reasoning. GPT-5.5 focuses on execution reliability over multiple turns rather than single-turn benchmark performance, per the Azure announcement.

The announcement placed particular emphasis on Foundry Agent Service as the operational layer. Microsoft has been positioning Foundry as the place where frontier models become governable production systems, separating the model access question from the deployment governance question. The April 24 GA coincided with broader Microsoft Build conference activity.

Third-party developer coverage noted the improved agentic coding specifically. Per the Azure blog, GPT-5.5 is designed to hold context across large codebases, diagnose ambiguous failures at the architectural level, and anticipate downstream testing and review requirements. Those remain vendor capability claims. Third-party benchmark comparisons against competing models had not been published at the time of the Microsoft announcement.

The Foundry platform also confirms compatibility with third-party agent frameworks including LangGraph and Claude Agent SDK alongside Microsoft's own Agent Framework and GitHub Copilot SDK. That cross-framework compatibility means GPT-5.5 is usable via Foundry Agent Service regardless of which orchestration layer a team has already adopted.

What to watch next

Two milestones are worth tracking. First, whether GPT-5.5 Pro becomes the default reasoning backend for GitHub Copilot Workspace's agentic tasks, given Microsoft's previously announced Project Polaris roadmap for Copilot model transitions. That roadmap targeted August 2026 for a Copilot model change in GitHub's agentic product. Second, whether third-party benchmarks from outside Microsoft and OpenAI confirm the computer-use accuracy and token efficiency claims before enterprise buyers commit at the Pro pricing tier. Any published evals on the computer-use dimension from research teams not affiliated with the two vendors will be worth close attention.

Sources