Provider · Creator · Some open weights

Microsoft models and API pricing

9 modelsHugging FacePrices checked 18 Aug 2026

Microsoft has 9 models in our catalogue. Input prices run from $0.07 per 1M tokens (Phi 4) to $0.62 (WizardLM-2 8x22B).

Ask an AI about this pageGChatGPTCClaudePPerplexityGeGeminiXGrok
More
Cite

Cite this page

ModelsAtlas. “Microsoft models and API pricing”. Retrieved 18 Aug 2026. https://modelsatlas.com/providers/microsoft
Models
9

in our catalogue

Input from
$0.07
observed We measured
Source
Our catalogue
Retrieved
17 Aug 2026

Phi 4

Max context
66K

WizardLM-2 8x22B

Tiers
1

Standard

Microsoft model line-up

Full pricing by tier →

Alphabetical · standard tier

ModelAtlasTiersInput / 1MOutput / 1MContext
deberta v3 baseMicrosoft · open weights—No live price——512
Florence 2 baseMicrosoft · open weights—No live price———
mdeberta v3 baseMicrosoft · open weights—No live price——512
phi 2Microsoft · open weights—No live price——2K
Phi 4MicrosoftGRIDB1standard $0.07$0.1416KAlternatives
table transformer detectionMicrosoft · open weights—No live price——1K
table transformer structure recognitionMicrosoft · open weights—No live price——1K
TRELLIS image largeMicrosoft · open weights—No live price———
WizardLM-2 8x22BMicrosoftGRIDD2standard $0.62$0.6266KAlternatives

Compare Microsoft with

Questions

What is the cheapest Microsoft model?
Phi 4 at $0.07 per 1M input tokens and $0.14 per 1M output on the standard tier.
Does Microsoft offer batch pricing?
No batch tier is listed for Microsoft models in our catalogue.
Does Microsoft have a free API tier?
No Microsoft model is listed at no per-token charge.
Which Microsoft model has the largest context window?
WizardLM-2 8x22B, at 65,535 tokens.

Sources

  • azure.microsoft.com — provider detailsAsserted · 16 Aug 2026
  • Our catalogue — prices and tiersObserved · 17 Aug 2026
  • Our catalogue — context windows and model recordsAsserted · 18 Aug 2026

Spotted something wrong? Report a correction · Methodology

Keep going in your AI

Opens with this page as the source. We add nothing else to the prompt.