Pricing reference · Australia · 2026
AZURE OPENAI PRICING AUSTRALIA.
Token rates by model, the difference between standard and provisioned throughput, AU region availability, and the patterns that decide what an Azure OpenAI deployment actually costs an Australian business.
Token rates by model — April 2026
Standard pay-as-you-go pricing.
USD list prices. Australian businesses are billed in AUD through the Azure subscription at Microsoft's regional conversion. Add GST for non-input-credit customers. Final invoiced AUD depends on currency at billing date and the contract type (CSP, EA, MCA).
| Model | Input | Output | Note |
|---|---|---|---|
| GPT-5 (standard) | USD $5.00 / 1M tokens | USD $15.00 / 1M tokens | General-purpose enterprise workhorse |
| GPT-5.5 (standard) | USD $5.00 / 1M tokens | USD $30.00 / 1M tokens | Strongest agentic + coding model. 1M context. Same per-token latency as 5.4. |
| GPT-5.4 Thinking (standard) | USD $4.00 / 1M tokens | USD $20.00 / 1M tokens | Deep reasoning; higher cost per call but fewer iterations |
| GPT-4o (standard) | USD $2.50 / 1M tokens | USD $10.00 / 1M tokens | Multimodal, faster, cheaper. Right model for many production workloads. |
| GPT-4o-mini (standard) | USD $0.15 / 1M tokens | USD $0.60 / 1M tokens | Lightweight tasks, classification, routing. Default for high-volume cheap work. |
| Embedding models (standard) | USD $0.02 to $0.13 / 1M tokens | n/a | Text embeddings for vector search and grounding |
Prices are Microsoft's published Azure OpenAI rates as of late April 2026. Rates change as new model versions ship — always verify against the live Azure pricing page before committing to a workload budget.
Standard vs Provisioned Throughput
When to switch.
Standard pay-as-you-go pricing is the right default for development, pilots, low-volume agents and anything where the daily token consumption is unpredictable. Provisioned Throughput Units reserve dedicated capacity at a flat hourly rate, with predictable latency and no rate limits.
The crossover for most Australian workloads is around 10 million tokens per day on a single model. Above that, PTUs are usually materially cheaper than standard rates and the latency profile tightens. Below that, standard pricing is simpler and cheaper. PTUs require a 1-month minimum commitment and are reserved at the hourly rate whether the workload uses them or not.
What an AU Azure OpenAI deployment usually costs
Three rough cost shapes Frontrow sees.
- Pilot / first agent: $200 to $2,000 per month. A focused Copilot Studio or Foundry agent grounded on a few hundred internal documents, used by a pilot group of 20 to 50 users. Standard pricing on GPT-4o or GPT-5 with selective routing to mini for lighter calls.
- Production agent at mid-market scale: $2,000 to $10,000 per month. A live agent serving hundreds to low thousands of interactions per day across a single business unit, with vector grounding, audit logging and a maintained prompt library. Standard pricing usually still wins.
- Production agent at enterprise scale: $5,000 to $20,000+ per month. Thousands to tens of thousands of interactions per day, often across multiple agents on shared models. Provisioned Throughput is usually the right pricing model. The cost predictability matters as much as the dollar figure for the CFO.
FAQ — Azure OpenAI pricing Australia
The questions Australian buyers ask first.
- What models are available in Azure OpenAI in 2026?
- The Azure OpenAI catalogue tracks the OpenAI frontier closely. As of April 2026 the catalogue includes the GPT-5 family (GPT-5, GPT-5.1, GPT-5.2, GPT-5.4 Thinking) and GPT-5.5 rolling through Microsoft Foundry, the GPT-4 family for compatibility, the o-series reasoning models, and the embedding and image models. Microsoft also hosts a curated set of partner models in Foundry, including Anthropic Claude variants and Meta Llama. Region availability differs by model — the latest frontier model is not always immediately in the Australia East region.
- Is Azure OpenAI available in the Australian region?
- Yes. The core Azure OpenAI services are deployed in Australia East. Microsoft committed AUD $25 billion in April 2026 to expand AU AI infrastructure, including significant additional capacity for Foundry workloads. Region-by-region model availability still varies and the latest models often launch in US regions before reaching Australia East — the lag is typically weeks, not months.
- How is Azure OpenAI priced?
- Two main pricing models. Standard (pay-as-you-go) charges per million input and output tokens for the model called, billed in the Azure subscription. Provisioned throughput units (PTUs) reserve dedicated capacity at a flat hourly rate, which is materially cheaper at scale and gives predictable performance for production workloads. Most Australian businesses start on standard pricing for development and pilots, then move to PTUs for production workloads above roughly 10 million tokens per day.
- What does Azure OpenAI typically cost for an Australian business?
- Highly dependent on the workload. A typical Australian mid-market Copilot Studio agent grounded on internal documents runs $200 to $2,000 per month on standard pricing for the model calls, plus the storage and compute for the grounding index. A production Foundry agent serving thousands of interactions per day usually moves to PTU at $5,000 to $20,000 per month for the dedicated capacity. The economics are driven by token volume and model selection — using a frontier reasoning model on every call when a faster cheaper model would do is the most common cost mistake Frontrow sees.
- How does Azure OpenAI compare to using OpenAI's API directly?
- Same models, same intelligence. Azure OpenAI adds enterprise grounding (Azure AD/Entra integration, Azure VNet, regional data residency), the Microsoft commercial agreement, AUD billing through the Azure subscription, and the Microsoft Purview integration for governance and audit. For an Australian business already on Azure, it removes the need for a second vendor relationship, USD billing rail, and parallel governance plumbing.
Walk an Azure OpenAI workload through with an Australian Microsoft Partner.
30-minute call. Model selection, region considerations, standard versus PTU economics, governance and the question of whether the workload should sit in Foundry or Copilot Studio.