Technology / WORLD

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.

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

Reading sizeSources ↓Share story ↗
Conceptual illustration of technology, processors and connected information.Illustration
Symbolic illustration

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

Timeline

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

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

  1. · Occurred

    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.

Methodology

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

Updated:

Table

Reported training configuration

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

Total parameters2800000000000
Active parameters80000000000
GB200 GPUs used16000
Methodology

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

Updated:

Structured coverage

Subjects and places

Subject proposals

Geographic references

Only source-provided geographic references are shown. Opening an external map is optional. Optional story maps use OpenStreetMap. Opening a map sends your IP address and this site’s address to the map provider; location data remains readable as text without loading the map.

  • United States · Place role: Coverage · Place type: Country or area · Location precision: Regional
    Evidence
    • Explicit preserved editorial country/region; no city or national centroid inferredmicrosoft.ai ↗
Taxonomy version: IPTC Media Topics 2026-Q2

Continue reading

Technology

AWS reports MoE kernel optimization results on SageMaker

AWS detailed GPU kernel optimizations for mixture-of-experts models on SageMaker AI on October 13. The company described swapping matrix operands, routing FP8 operations to the Triton backend and removing unnecessary just-in-time recompilations.

AWS Artificial Intelligence Blog1 min read
Data included