Microsoft AI: MAI-Base-2 pricing and API availability
Official sources verified . Access and price status are reviewed daily.
Pricing and availability
| Item | Verified MAI-Base-2 status |
|---|---|
| Input / output per million tokens | Not published; unknown, not free |
| Public model ID / endpoint | Not announced |
| Model development | post-training |
| Weights and licensing | Future intent to license to trusted Foundry partners; no public download, license terms or fee announced |
Training efficiency does not establish an inference price
Microsoft reports 2.8 trillion total and 80 billion active parameters, with 16,000 GB200 GPUs over 38 days. Its 4.2–12.3x pre-training FLOPs-efficiency estimate against DeepSeek-V4-Pro-Base is a matched-evaluation-loss comparison. Microsoft says it spent the saving on a larger model. Neither the ratio nor GPU-days determine an API bill.
See our MAI-Base-2 announcement and buyer analysis. Do not apply another MAI model, Azure OpenAI, Microsoft 365 Copilot, or a Foundry hosting rate to this unreleased product.
Labs availability blocker
Still undergoing post-training; no public exact-model route, weights, token rate or authorized Labs access verified. Vendor-held-out BPB comparisons are not Labs results. A fair result needs authorized access, a pinned model and serving configuration, fixed tasks, raw outputs, latency, retries, complete token counts and billed or allocated serving costs. Read our Labs coverage notes.
Choose an available route today
Compare callable alternatives on the API pricing table and token calculator. For managed open-model alternatives, see Novita's current catalog; this does not imply MAI-Base-2 availability. We may earn a commission from that sponsored link.
Official sources and daily maintenance
- microsoft.ai/news/frontier-pre-training-at-microsoft-ai/
- azure.microsoft.com/en-us/pricing/details/ai-foundry-models/microsoft/
- ai.azure.com/catalog
The canonical availability record and API mirror publish null rates for unknown prices, outside the priced-model roster. Daily official-source and freshness monitoring flags changes for Data review and paired republishing; verification dates are not blindly advanced. Existing priced API rows are unchanged.
Frequently asked questions
What is the MAI-Base-2 API price?
Microsoft has not published a MAI-Base-2 input or output token rate in the reviewed announcement, Microsoft model pricing page or public catalog. Unknown does not mean free.
Can I use MAI-Base-2 now?
Microsoft says MAI-Base-2 is undergoing post-training. No public exact-model endpoint, API ID, downloadable weights or licensing fee is verified.
Does 4–12x training efficiency mean cheaper tokens?
No. The claim concerns estimated pre-training FLOPs at matched held-out evaluation loss. It is not an inference speedup or token-price cut.