microsoft
Microsoft: Phi 4
microsoft/phi-4
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
microsoft/phi-4 is a 14-billion-parameter dense decoder-only Transformer from Microsoft Research, released under the MIT license and served as a text-to-text model with a 16K-token context window. Microsoft describes it as aimed at complex reasoning, math, and code while fitting memory-constrained and latency-sensitive deployments. Vendor-run evaluations covered benchmarks including MMLU, MATH, GPQA, and HumanEval. Training was primarily in English, with multilingual data around 8 percent. Evidence here is a vendor model card plus an aggregator catalog, so workload-specific validation remains advisable.[1][2]
Strengths
- Vendor-described focus on complex reasoning, math, and code, trained on 9.8T tokens and aligned with supervised fine-tuning plus direct preference optimization.[1][2]
- Text-to-text model with a 16K-token context window, positioned by its authors for memory- or compute-constrained and latency-bound settings.[1][2]
- OpenRouter's catalog lists support for structured outputs and common sampling controls such as temperature, top_p, top_k, and seed.[1]
Limitations
- Trained primarily for English: multilingual data was about 8 percent of training, and the vendor says other languages and less-represented English varieties may perform worse.[2]
- The vendor warns of possible inaccurate or fabricated outputs, biased or offensive content, and narrow code coverage centered on Python with common packages; knowledge cutoff is June 2024.[1][2]
- Evidence is limited to the vendor model card and OpenRouter's catalog; no independent evaluations were in this source pack, so validate accuracy and safety on your own workload.[1][2]
Where to use it
- Reasoning-, math-, or logic-heavy assistant features that need a mid-size model; treat as a suggestion to test, not a proven fit.[1][2]
- Python-centric code assistance where generated API calls are manually verified, per the vendor's own guidance.[2]
- English-language chat or instruction-following prototypes in resource- or latency-constrained environments, with application-level safeguards added before high-risk use.[1][2]
Sources
- [1] OpenRouter catalog · Microsoft: Phi 4Sources checked: Nov 13, 2026
- [2] Model documentation · Microsoft: Phi 4Sources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
[Microsoft Research](/microsoft) Phi-4 is designed to perform well in complex reasoning tasks and can operate efficiently in situations with limited memory or where quick responses are needed. At 14 billion...
Specifications
- Model ID
- microsoft/phi-4
- Creator
- microsoft
- Added to OpenRouter
- 10.01.2025
- Context length
- 16,384
- Modality
- text->text
- Input modalities
- text
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.070 / out $0.140
- Knowledge cutoff
- 30.06.2024
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | No measurement | — |
| Coding | No measurement | — |
| Agents | No measurement | parameters: structured_outputs |
| Writing | No measurement | output: text |
| Reasoning | No measurement | — |
| Document analysis | No measurement | — |
| Vision | No measurement | — |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | — |
| Embeddings | No measurement | — |
Evaluations
No evaluations are supplied for this model.
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| DeepInfrabf16 | 16,384 | $0.056 | $0.112 | — | — | — |
| DeepInfrabf16 | 16,384 | $0.070 | $0.140 | — | — | 100% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.