Phi-4

Available · Language · Milestone

Phi-4 is a 14 billion parameter small language model from Microsoft, released on 12 December 2024 and aimed at complex reasoning such as math as well as conventional language tasks. Microsoft offers it on Azure AI Foundry and as open weights on Hugging Face under the MIT License; its report describes a training recipe centred on data quality, with synthetic data used throughout. [1] [2] [3] Primary source

Timeline of Phi-4 →

Claims and evidence

  • Released Primary source[1]page date; Today we are introducing Phi-4 [2]Model Summary, Release date [3]Submission history, v1
  • Status AvailableWeights downloadable from Microsoft's Hugging Face organisation and listed as GA, with no retirement date, on the Microsoft Foundry model retirement schedule (checked 2026-10-01). Primary source[2] [4]Microsoft table: Phi-4, GA
  • Successor of Phi-3.5 EditorialEditorial link along the core Phi line: the post calls Phi-4 the latest member of the Phi family and the report describes minimal changes to the phi-3 architecture, without naming a direct predecessor; Phi-3.5 was the most recent core text release before it. Primary source[1]first paragraph [3]abstract
  • Change · Training data Training data extends the Phi-3 data with new synthetic textbook-like data and acquired academic books and Q&A datasets; synthetic data is used throughout.Compared with Phi-3.5 Primary source[2] [3]
  • Change · Context length Context length is 16K tokens, compared with 128K tokens for the Phi-3.5 models.Compared with Phi-3.5 Primary source[2] [5]
  • Change · Other Aimed at complex reasoning such as math; the report credits data quality, the training curriculum and post-training rather than architecture changes.Compared with Phi-3.5 Primary source[3] [1]
  • Input text Primary source[2]Model Summary, Inputs
  • Output text Primary source[2]Model Summary, Outputs
  • Open weights Yes Primary source[2]Model Summary, License
  • Context window 16K tokens tokens Primary source[2]Model Summary, Context length
  • Parameters 14B Primary source[1] [3]abstract [2]Model Summary, Architecture
  • Access API, Open-weights downloadThe api value stands for the Azure AI Foundry offering named in the announcement. Primary source[1] [2]

Lineage

Predecessors

Successors

No known successor.

Based on

Not derived from another model.

Variants and derived

Siblings

None recorded.

All ancestors

All descendants

Variants

No variants recorded in this record.

Related AI Radar coverage

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

All model releases from Microsoft on AI Radar →