AIGROW vs Microsoft Foundry

Microsoft Foundry, the Azure service that includes Azure OpenAI, sells GPT and Grok itself and Claude through Microsoft Marketplace, at list price on Global deployments and 10% more on Data Zone ones, billed to an Azure subscription. AIGROW sells discounted prepaid credit for the same models through an OpenAI-compatible endpoint, with no Azure subscription. The table compares the models both sell, price by price.

See the table

Every model both sell costs less per token through AIGROW credit than at Microsoft Foundry, on input and on output.

9AIGROW vs Microsoft Foundrymodels both sell, priced side by side
Less through AIGROW credit
9 of 9
Microsoft Foundry prices read
18 November 2026
Model check
11 November 2026, 15 November 2026, 16 November 2026 and 18 November 2026
Sponsored placement
none
Nothing on this page is paid for by anyone.

Price per model: Microsoft Foundry and AIGROW credit

USD per million tokens, input on the first line and output on the second. The difference is rounded so it never overstates a saving.

Prices per million tokens at Microsoft Foundry and through AIGROW credit
ModelMicrosoft FoundryAIGROW creditDifference
GPT-6.1 Terra$1.00$5.00$0.25$1.2575% less75% less
GPT-6 Luna$0.10$0.50$0.025$0.12575% less75% less
Grok 4.7$2.00$6.00$0.50$1.5075% less75% less
Claude Opus 5.5$4.00$20.00$1.60$8.0060% less60% less
Claude Sonnet 5.5$2.00$10.00$0.80$4.0060% less60% less
Claude Haiku 5.5$0.10$0.50$0.04$0.2060% less60% less
GPT-6 AstraPrevious version, replaced by GPT-6.2 Astra$10.00$50.00$2.50$12.5075% less75% less
GPT-6.1 SolPrevious version, replaced by GPT-6.2 Sol$2.00$10.00$0.50$2.5075% less75% less
Claude Fable 5.1Previous version, replaced by Claude Fable 5.2$10.00$50.00$4.00$20.0060% less60% less

Microsoft Foundry’s prices as published on 18 November 2026, each linked from its row. AIGROW’s are the rates the endpoint bills with, at the price credit sells for. Prices change, so check theirs before you decide.

Microsoft’s prices here are Global Standard deployments at short context on 18 November 2026: GPT and Grok as Azure’s price list gives them, Claude as Anthropic’s offers on Microsoft Marketplace give them. Data Zone deployments cost 10% more. GPT, Grok and Claude Haiku have higher long-context rates. Microsoft also prints a Flex rate at half price for some GPT models and a batch rate at 50% off, which AIGROW does not have.

Fees and terms

What each one charges beyond the tokens, and what happens to credit you have not spent.

TermsAIGROWMicrosoft Foundry
Fee when you buy creditNone: $10 buys $40 of creditNone on Global deployments; Data Zone deployments cost 10% more
Smallest purchase$10None: pay-as-you-go has no upfront commitment
Credit expiryNo expiry date in our TermsNo prepaid credit to expire: usage is paid in arrears
RefundsSpent credit: never. Unspent credit: within 14 days of buying it, for consumers in the EU, EEA, UK and Switzerland.Not stated for pay-as-you-go usage on the pages we read
PaymentCard, Apple Pay, Google Pay or Amazon Pay at the Stripe checkout; WeChat Pay, Pix, UPI or crypto for a one-off packThe payment method on the Azure subscription, for the preceding month’s usage
Models17 models from Anthropic, OpenAI, Google and xAIOpenAI, xAI and other models sold by Azure; Claude and more from partners

Microsoft Foundry’s terms as published on the pages below, read on 18 November 2026. AIGROW’s are from our Terms of Service.

Azure OpenAI pricing · Microsoft on deployment types · Microsoft on Claude in Foundry · Microsoft on data and privacy

Is the model the one you paid for?

How each side shows that the model answering is the one it sells.

Microsoft Foundry
Microsoft says prompts and completions sent to the models Azure sells are not available to OpenAI or the other makers, and it offers Claude in two versions, one hosted on Azure and one hosted on Anthropic’s infrastructure. As of 18 November 2026 we have not run our check on Microsoft Foundry. The script below can produce a result for either side.
Certified genuine by AIGROW
On 11 November 2026, 15 November 2026, 16 November 2026 and 18 November 2026 we sent each model we sell the same prompts as the same model at its maker or through OpenRouter, and compared the answers, the prompt tokens and the tool calls. 9 of the 9 models on this page met our rule. This is AIGROW’s own check: no model maker reviewed or endorsed it. See every result and the method
Results of the model check of 11 November 2026, 15 November 2026, 16 November 2026 and 18 November 2026
ModelHereAt the maker
Compared with Anthropic’s API
Claude Opus 5.512 of 1212 of 12
Claude Sonnet 5.512 of 1212 of 12
Claude Haiku 5.512 of 1212 of 12
Claude Fable 5.112 of 1212 of 12
Compared with the same model through OpenRouter
GPT-6.1 Terra11 of 1212 of 12 with effort set to high12 of 1211 of 12 with effort set to high
GPT-6 Luna8 of 128 of 12 with effort set to high9 of 1210 of 12 with effort set to high
Grok 4.75 of 55 of 5
GPT-6 Astra12 of 1212 of 12
GPT-6.1 Sol12 of 1212 of 12

A 2026 study by the CISPA Helmholtz Center for Information Security tested 17 unofficial APIs that claim to give access to the makers’ models, and found identity verification failures in 45.83% of its fingerprint tests. It left out routing platforms that disclose where they send requests, such as OpenRouter. Any seller below list price can be checked with the script below, AIGROW included. Read the study.

Run the check yourself

Tests a GPT-6.1 Sol deployment on Azure beside the same model through OpenRouter, with the Azure resource’s API key and an OpenRouter key as the reference. Azure addresses a model by the deployment name you chose. To test AIGROW the same way, leave out --base and MAKER_MODEL, use an AIGROW key and name the model gpt-6.1-sol.

MODEL_CHECK_KEY=your_azure_api_key MAKER_KEY=sk-or-your_key MAKER_MODEL=openai/gpt-6.1-sol node model-check.mjs --base https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1 your-deployment-name

Download model-check.mjs

Every request is billed as usual. The battery is the costly one: a few cents of credit on most models, more when a model thinks for long. To test another seller’s API, pass its address with --base.

When Microsoft Foundry is the better choice

Inside your Azure subscription
Usage is charged to an Azure subscription and requests can be signed with Microsoft Entra ID as well as an API key. AIGROW is a separate prepaid account with its own keys.
Where the data is processed
Foundry lets you choose global, data zone or single-geography processing for a deployment, and sells reserved capacity and batch processing. AIGROW has one rate per model and no regional endpoint.
GPT served by Microsoft
Microsoft says prompts and completions sent to the models Azure sells are not available to OpenAI and are not used to train foundation models without your permission. AIGROW is an independent service outside Azure, and its Privacy Policy says how requests are handled.

Switching from Microsoft Foundry

A base URL and a key that starts with ag_, created in API access on the dashboard.

  • At Microsoft Foundry today: base URL https://YOUR-RESOURCE-NAME.openai.azure.com/openai/v1
  • Anthropic SDK and Claude Code: base URL https://aigrow.gg
  • OpenAI SDK, Cline, Aider, OpenCode: base URL https://aigrow.gg/v1

Some lines behave differently from the maker’s endpoint, for example on max_tokens and on the starting reasoning effort: the model check lists each difference with its numbers. Codex sends Responses API requests, which this endpoint does not take.

The setup guide covers every other tool, and the Claude Code page walks through its settings one by one.

For the models in the table, Global Standard deployments bill list price and Data Zone deployments 10% more. The table sets each model’s price through AIGROW credit beside it, with the difference worked out from the rates our endpoint bills with. Microsoft prints batch processing at 50% off its Global Standard price, which AIGROW does not offer.

Yes. Microsoft’s documentation now calls it Azure OpenAI in Microsoft Foundry Models, and the same service sells Grok and other models. Claude is deployed in Foundry too, through a subscription to Anthropic’s offer on Microsoft Marketplace.

Microsoft hosts the models Azure sells and offers Claude hosted on Azure or on Anthropic’s infrastructure. Run the same test on both. Our public script sends any endpoint twelve questions with one right answer, a two-word prompt, four short questions and a tool call, and prints the score and the prompt tokens beside the same model at its maker or through OpenRouter. We have not run it on Microsoft Foundry.

Those that send OpenAI Chat Completions or Anthropic Messages requests do: set the base URL https://aigrow.gg/v1, use an AIGROW key and name the model by its id, such as gpt-6.1-sol, where Azure takes your deployment name. AIGROW has no Responses API endpoint, so code that calls it has to switch to Chat Completions.

No. Microsoft’s subscription agreement says pay-as-you-go customers pay for the preceding month’s usage with no upfront commitment, so there is no balance to top up or to expire. AIGROW is prepaid: you buy credit and requests draw on it. AIGROW’s Terms set no expiry date for credit. Spent credit is not refunded, and consumers in the EU, EEA, UK and Switzerland can have unspent credit refunded within 14 days of buying it.

No. AIGROW is independent and is not affiliated with, endorsed by or sponsored by Microsoft. The Azure and Microsoft Foundry names identify Microsoft’s products only.

Run your own workload on both, then compare the bills.

Start with the $10 pack, run your real workload against it, and check the dashboard before buying more.

Monthly plans or one-off packs · redemption code delivered at once · usage dashboard included