AI Radar

Phi-3.5

Available · Language

Phi-3.5 is a Microsoft release of two open-weight text models announced on 22 August 2024: Phi-3.5-mini (3.8 billion parameters) and the mixture-of-experts Phi-3.5-MoE (42 billion in total, 6.6 billion active). Both support multiple languages and 128K tokens of context; Phi-3.5-mini was also offered on a serverless endpoint in Azure AI Studio. [1] [2] [3] Primary source

Timeline of Phi-3.5 →

Claims and evidence

  • Announced Primary source[1]page date; Today, we are proud to announce
  • Released Primary source[2]Model, Release date [3]Model, Release date
  • Retired from Microsoft Foundry Primary source[4]Microsoft table
  • Status AvailableWeights of both variants are still downloadable from Microsoft's Hugging Face organisation (checked 2026-10-01). Retired from Microsoft Foundry on 2025-08-30 (see milestones). Primary source[2] [3]
  • Successor of Phi-3The Phi-3.5-mini card calls it an update over the June 2024 Phi-3 Mini release; the post presents Phi-3.5 as the response to feedback since the April 2024 Phi-3 launch. Primary source[2]Release Notes [1]introduction
  • Change · Languages Adds multilingual support: the Phi-3.5-mini card lists 23 languages, and Microsoft says Phi-3.5-MoE supports more than 20.Compared with Phi-3 Primary source[2] [1]
  • Change · Architecture Adds a mixture-of-experts model, Phi-3.5-MoE, with 16 experts of 3.8 billion parameters each; 6.6 billion of its 42 billion parameters are active.Compared with Phi-3 Primary source[1] [3]
  • Change · Context length Phi-3.5-mini comes only in a 128K-token version, where Phi-3-mini also had a 4K version; the report describes LongRoPE to extend the context.Compared with Phi-3 Primary source[5] [2] [6]
  • Change · Training data Phi-3.5-mini adds continued pre-training on multilingual synthetic and filtered data, and proximal policy optimization in post-training.Compared with Phi-3 Primary source[1]
  • Input text Primary source[2]Model, Inputs [3]Model, Inputs
  • Output text Primary source[2]Model, Outputs [3]Model, Outputs
  • Feature Multilingual Primary source[2]Supported languages [3]Supported languages [1]
  • Open weights Yes Primary source[2]License [3]License
  • Context window 128K tokens Primary source[1] [2]Model, Context length [3]Model, Context length
  • Access API, Open-weights downloadThe api value rests on the Phi-3.5-mini serverless endpoint in Azure AI Studio. Primary source[1]More Predictable Outputs: Phi-3.5-mini serverless endpoint in Azure AI Studio [2]

Lineage

Predecessors

Successors

Based on

Not derived from another model.

Variants and derived

Siblings

None recorded.

All ancestors

All descendants

Variants

Phi-3.5-mini

Same dates as Phi-3.5.

  • Variant Inline variant in this record. Primary source[1] [2]

Also known as: Phi-3.5-mini-instruct

Differs in:

  • Parameters: 3.8B Evidence not assessed [2]Model, Architecture [1]Phi-3.5-mini

Phi-3.5-MoE

Same dates as Phi-3.5.

  • Variant Inline variant in this record.Mixture-of-experts model; Microsoft called it a new member of the Phi family. Primary source[1] [3]

Also known as: Phi-3.5-MoE-instruct

Differs in:

  • Parameters: 42B total (16x3.8B experts), 6.6B active Evidence not assessed [1]Phi-3.5-MoE: Mixture-of-Experts [5]v4, phi-3.5-MoE architecture [3]Model, Architecture

Related AI Radar coverage

AI Radar coverage starts in June 2026; no coverage linked yet.

All model releases from Microsoft on AI Radar →