# Microsoft reports completion of MAI-Base-2 pre-training

> Microsoft AI said on October 15 that it completed pre-training MAI-Base-2, a mixture-of-experts model with 2.8 trillion total and 80 billion active parameters. The company says training used 16,000 GB200 GPUs over 38 days in August and September; the model is still undergoing post-training.

- Canonical URL: https://newsverum.com/en/news/microsoft-mai-base-two-pretraining-report-october-2026
- Publisher: Newsverum
- Section: Technology
- Country: United States
- Published: 2026-10-15T20:19:53+00:00
- Updated: 2026-10-15T20:19:53+00:00
- Source review period: 2026-10-15
- Other editions: [Türkçe](https://newsverum.com/tr/news/microsoft-mai-base-two-pretraining-report-october-2026.md), [العربية](https://newsverum.com/ar/news/microsoft-mai-base-two-pretraining-report-october-2026.md), [Français](https://newsverum.com/fr/news/microsoft-mai-base-two-pretraining-report-october-2026.md), [Español](https://newsverum.com/es/news/microsoft-mai-base-two-pretraining-report-october-2026.md), [Deutsch](https://newsverum.com/de/news/microsoft-mai-base-two-pretraining-report-october-2026.md), [Português](https://newsverum.com/pt/news/microsoft-mai-base-two-pretraining-report-october-2026.md)

*The source does not state the event date; the period shown is the source review period.*

Microsoft AI said on October 15 that it completed pre-training MAI-Base-2, a mixture-of-experts model with 2.8 trillion total and 80 billion active parameters. The company says training used 16,000 GB200 GPUs over 38 days in August and September; the model is still undergoing post-training.

Using its own held-out evaluations and scaling fits, Microsoft estimates 4.2–12.3 times greater compute efficiency to match DeepSeek-V4-Pro-Base. This is not an independent product-speed test; licensing weights to trusted partners through Foundry was described as a future plan.

## Data

### Microsoft reports completion of MAI-Base-2 pre-training

Microsoft reports completion of MAI-Base-2 pre-training

Methodology: Reported research results are attributed to the source; no independent performance measurement was made.

| date | label | description | status |
|---|---|---|---|
| 2026-10-15 | Publication of the research report | Microsoft AI said on October 15 that it completed pre-training MAI-Base-2, a mixture-of-experts model with 2.8 trillion total and 80 billion active parameters. The company says training used 16,000 GB200 GPUs over 38 days in August and September; the model is still undergoing post-training. | occurred |

Sources: <https://microsoft.ai/news/frontier-pre-training-at-microsoft-ai/> · Download data: [JSON](https://newsverum.com/api/v1/articles/microsoft-mai-base-two-pretraining-report-october-2026/data/research-publication?lang=en&format=json) · [CSV](https://newsverum.com/api/v1/articles/microsoft-mai-base-two-pretraining-report-october-2026/data/research-publication?lang=en&format=csv)

### Reported training configuration

Microsoft reports completion of MAI-Base-2 pre-training

Methodology: Reported research results are attributed to the source; no independent performance measurement was made.

| Measure | Count |
|---|---|
| Total parameters | 2800000000000 |
| Active parameters | 80000000000 |
| GB200 GPUs used | 16000 |

Sources: <https://microsoft.ai/news/frontier-pre-training-at-microsoft-ai/> · Download data: [JSON](https://newsverum.com/api/v1/articles/microsoft-mai-base-two-pretraining-report-october-2026/data/reported-training-configuration?lang=en&format=json) · [CSV](https://newsverum.com/api/v1/articles/microsoft-mai-base-two-pretraining-report-october-2026/data/reported-training-configuration?lang=en&format=csv)

## Sources

- [Microsoft AI](<https://microsoft.ai/news/frontier-pre-training-at-microsoft-ai/>) (Source checked: 2026-10-15; Published: 2026-10-15)

---

Cite the canonical URL above. [Editorial standards and corrections](https://newsverum.com/en/editorial)