seminal.ai
← All open-weight models

Microsoft · General language

Phi-3.5-MoE-instruct

The largest Phi ever shipped (16x3.8B MoE, 6.6B active) — historically interesting and still self-hostable under MIT, but superseded on quality-per-dollar by Phi-4 and removed from the current Foundry model list.

Open weights · Permissive (Apache, MIT, OpenMDW) · Data checked September 6, 2026 · Not a hands-on evaluation

Model ID Phi-3.5-MoE-instruct

Parameters
41.9B total · 6.6B active41.9B-A6.6B MoE
Architecture
Mixture of experts
Context window
131,072 tokensMaximum input
Max output
4,096 tokensPer response
Input price
$0.16 per 1M tokens
Output price
$0.64 per 1M tokens
Cached input
Not published
Blended price
$0.28 per 1M tokens3:1 input to output
Knowledge cutoff
2023-10
Released
2024-08-17
Status
deprecated
Size band
40B to 120B

License and openness

Phi-3.5-MoE-instruct is released under the MIT. This is a permissive license. It generally allows use, modification and redistribution, including commercial use, subject to attribution and notice requirements. License terms can change between versions; confirm the text published with the weights you download.

Capabilities

Input modalitiestext
Output modalitiestext

What it is for

At a three-to-one input-to-output ratio, Phi-3.5-MoE-instruct costs $0.28 per million tokens blended. A workload of one million input and 330,000 output tokens per day would run about $11.14 per month at list price, before caching or batch discounts.

Cost at list price

Input$0.16 per 1M tokens
Output$0.64 per 1M tokens
Cached inputNot published
Blended, 3:1$0.28 per 1M tokens
Estimate a workload across all models ↗

Sources

Sources checked September 6, 2026. Specifications and prices change without notice; confirm against the provider before you commit.

Other open-weight models from Microsoft

See all 23 from Microsoft ↗

Browse all open-weight models ↗