Azure AI Foundry vs GitHub Copilot vs One Gateway

Ask Microsoft how to call frontier models from your application and the answer is Azure AI Foundry. Ask how to get those same models into your developers' terminals and the answer is GitHub Copilot. Two products, two brands, two billing systems, two allowance mechanics — with overlapping catalogues, since OpenAI and Anthropic model families show up on both sides.

That split is the most consequential thing to understand in any Azure AI Foundry vs GitHub Copilot comparison, because the honest answer to "which one?" is that Microsoft expects most engineering organizations to buy both. And as of this month, a complete setup is actually three products, not two.

Azure AI Foundry vs GitHub Copilot: different products, overlapping models

The two products don't compete with each other. They partition the market by use case:

Since June 1, 2026, Copilot's chat, agent mode, code review, and CLI all meter as AI Credits at token-based rates — code review can additionally consume GitHub Actions minutes — so both products now bill you by the token for overlapping models. What you cannot do is share a dollar between them. A Copilot credit is useless to your production API traffic; your Azure commitment does not top up a developer who hit their Copilot allowance mid-refactor.

The third product nobody mentions in the sales deck

Foundry gives you model access, not governance. If you want shared rate limits, token quotas, and failover in front of those models, Microsoft's answer is a third product: in August 2026, Azure API Management shipped a dedicated AI Gateway tier in public preview — token and request limits, quotas, content safety, and model fallback, as its own APIM tier. Preview pricing has not been announced yet.

Microsoft's own architecture guidance is candid about the arithmetic: use APIM when you need shared policy, use Foundry's model router when you need prompt-aware model choice, and use both when you need governed traffic and smart selection. So the complete Microsoft stack for a team that ships an AI product and uses AI to build it is:

1. Azure AI Foundry — model access, billed through Azure
2. APIM AI Gateway tier — limits, quotas, fallback, its own APIM tier (preview)
3. GitHub Copilot — coding agents, billed per seat plus credits

Three products, two vendors' worth of dashboards, and three separate places a cost overrun can hide.

What the same two use cases cost on one platform

LLM Gateway sells both use cases as one platform with one model catalogue and one balance:

• The API use case — an OpenAI-compatible gateway in front of 200+ models from 40+ providers. Tokens at provider list price with a flat 5% fee on credits, or 0% with your own provider keys. Configurable budgets with hard caps per organization, project, and API key, prompt caching, and per-request cost attribution are built in — the things the APIM AI Gateway tier packages separately.
• The coding use case — DevPass flat plans for agents like Claude Code, Cursor, Cline, and OpenCode, drawing on the same catalogue.

The plan math is where the difference stops being philosophical. Compare what a dollar of subscription buys in metered usage:

Copilot plan prices and credit counts are from GitHub's published plans as of August 2026, at $0.01 per credit. DevPass allowances are metered at provider list rates, so a dollar of allowance buys exactly a dollar of tokens — the mechanics are in what is an AI coding plan.

Two things Copilot includes that this table doesn't capture: unlimited inline code completions on every paid plan, and the GitHub-native integration itself. If completions are most of what your team uses, Copilot's $10 tier is genuinely hard to beat. The gap opens with agent workloads — the token-hungry ones — where the allowance, not the seat price, is the real product.

You don't have to leave Azure to consolidate

The part that surprises Azure-committed teams: Azure OpenAI and Azure AI Foundry are built-in LLM Gateway providers. Bring your Azure credentials, and your traffic still runs on your Microsoft agreement and your negotiated rates — with 0% gateway markup. What changes is what sits in front of it: automatic failover to other providers when a deployment degrades, response caching, and one cost dashboard that covers your API traffic and your coding agents in the same currency. If your requirements say prompts must never transit a third party, the gateway is open source under AGPLv3 — run it inside your own VPC and keep the routing layer in your boundary too.

Consolidating the bill does not mean abandoning the infrastructure. It means the routing and governance layer stops being a Microsoft SKU decision. The full comparison is at LLM Gateway vs Azure AI Foundry.

Who should still buy the Microsoft stack

Honesty section. The three-product stack is the right call when:

• Your spend is already committed. If you have a Microsoft Azure Consumption Commitment to burn down, Foundry consumption counts against it and effectively discounts itself.
• You need PTU-guaranteed capacity. Provisioned throughput reservations give latency guarantees a multi-provider gateway routes around rather than replicates.
• Copilot's completions are the product you're buying. For inline-completion-heavy teams that live in VS Code and GitHub PRs, the $10–$19 seats are cheap and the integration is unmatched.
• Procurement wants one throat to choke. Three products from one vendor can be easier to sign than two vendors, even when it costs more.

If instead you're paying for both use cases separately, watching Copilot overages nobody capped, and running a third gateway tier just to get rate limits — the consolidation math is worth an afternoon. Migration guides exist for both directions of the stack: GitHub Copilot for the coding side, and the Azure AI Foundry comparison for the API side.

Run the numbers on your own stack

• Try LLM Gateway free — one API for 200+ models, budgets and caching included, BYOK at 0%
• Compare DevPass coding plans — 3x the plan price in usage at provider list rates, on every tier
• Weighing the developer-tool side alone? Read the best GitHub Copilot alternatives