Plain-English Summary
A Phi-4 variant tuned specifically for multi-step reasoning in math and complex problem solving.
Phi-4-reasoning
A Phi-4 variant tuned specifically for multi-step reasoning in math and complex problem solving.
Phi-4
Reasoning LLM
GA / open weights
Open weights: Yes
API: Yes and self-hosted
Technical Notes
Reasoning-tuned text model with 32K input and output limits and open weights.
Best For
Math, logic, reasoning research, private analytical assistants
Watch Out For
Reasoning can be verbose; less suitable for broad multimodal tasks.
Model Specs
- Maker
- Microsoft
- Country
- USA
- Family
- Phi-4
- Type
- Reasoning LLM
- Release date
- 2025-04-30
- Date confidence
- Exact
- Status
- GA / open weights
- Context window
- 32,768 tokens
- Max output
- 32,768
- Parameters
- 14B
- Active parameters
- 14B
- Architecture
- Dense reasoning transformer
- Reasoning
- Yes — specialized
- Tool calling
- Host dependent
- Structured output
- Host dependent
- API available
- Yes and self-hosted
- Open weights
- Yes
- Self-hostable
- Yes
- License
- MIT
- Inputs
- Text
- Outputs
- Text
- Approx. price
- See provider pricing
- How to access
- Provider website
- Model / API ID
- Phi-4-reasoning
- Knowledge cutoff
- Not publicly disclosed
- Fine-tuning
- Fine-tuning, quantization, self-hosting
- Completeness
- 100 %
- Verified
- Jul 18, 2026
- Snapshot date
- Jul 18, 2026
- Source
- Official source