Phi 4
Microsofthuggingface.co/microsoft/phi-4
[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...
Updated 32 min ago · first seen 11 Sept 2026
model_01M294WWM26ZA9FAW9BCD0QYCJ
- Parameters
- 14.7B
- T2 · 1 h ago
- Context
- 16.4K tokens
- T2 · 1 h ago
- Released
- 11 Dec 2024
- T2 · 33 min ago
- License
- mit
- T2 · 33 min ago
- Knowledge cutoff
- Jun 2024
- T2 · 1 h ago
Specification
- Release date
- 11 Dec 2024
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- Openness
- open-weights
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 1 h agomedium
- License
- mit
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- Architecture
- Phi3ForCausalLM
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- Parameters
- 14.7B
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 1 h agomedium
- Context window
- 16.4K tokens
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Max output
- 14.8K tokens
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Knowledge cutoff
- Jun 2024
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Modalities
- text
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Input modalities
- text
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Output modalities
- text
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Languages
- en
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- File size
- 29.3 GB
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- Hugging Face repo
- microsoft/phi-4
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Pipeline tag
- text-generation
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 1 h agomedium
- Model card
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 1 h agomedium
- Downloads
- 708,128
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 32 min agomedium
- Likes
- 2,299
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 32 min agomedium
- Gated
- No
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- Hf inference providers
- deepinfra, featherless-ai
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- Last modified
- 2026-07-14T14:22:25+00:00
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 1 h agomedium
- Library name
- transformers
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- License url
- https://huggingface.co/microsoft/phi-4/resolve/main/LICENSE
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- Downloads all time
- 14,247,419
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 32 min agomedium
- Model type
- phi3
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- Openrouter id
- microsoft/phi-4
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Supported parameters
- frequency_penalty, logit_bias, max_tokens, min_p, presence_penalty, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, top_k, top_p
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Tags
- transformers, safetensors, phi3, text-generation, phi, nlp, math, code, chat, en
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
- Weights dtype
- BF16
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
Each value shows its source, tier and observation time. Conflicting claims are kept side by side and flagged — never averaged. How AI Atlas records facts →
Provenance
Attributed facts
32
Source tiers
T232
Freshest observation
32 min ago
Conflicts
None
Modalities
- Modalities
- text
- Input
- text
- Output
- text
Capabilities
Tool calling
No
OpenRouter public model & pricing listing · T2
Structured output
Yes
OpenRouter public model & pricing listing · T2
Reasoning
Unavailable
Vision
Unavailable
Audio
Unavailable
Fine-tuning available
Unavailable
- Context window
- 16.4K tokens
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Max output
- 14.8K tokens
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Knowledge cutoff
- Jun 2024
Source:OpenRouter public model & pricing listingT2observed 1 h agomedium
- Languages
- en
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 33 min agomedium
No benchmark results recorded
Current prices
| Provider | Input / 1M | Output / 1M | Cached in | Batch in / out | Context | Observed | Source |
|---|---|---|---|---|---|---|---|
| OpenRoutermicrosoft/phi-4 | $0.07 | $0.14 | — | — | 16.4K | 1 h ago | openrouter.aiT2 |
USD per 1M tokens as published by each provider (USD). Rows are append-only: every change is kept in the history below.
Available through 1
Price history
Input price · USD / 1M tokens
Output price · USD / 1M tokens
- OpenRouter
Memory need = bytes per parameter (4-bit ≈ 0.5 × 1.15 overhead, 8-bit 1.0, fp16 2.0) + a KV-cache allowance. Not a measurement.
| Hardware | Quantization | Memory | Est. need | Fits |
|---|---|---|---|---|
| Apple M3 Ultra | 4bit | — | 8.9 GB | Yes |
| AMD Instinct MI325X | 4bit | 256 GB | 8.9 GB | Yes |
| AMD Instinct MI300X | 4bit | 192 GB | 8.9 GB | Yes |
| Apple M2 Ultra | 4bit | — | 8.9 GB | Yes |
| NVIDIA B200 | 4bit | 180 GB | 8.9 GB | Yes |
| NVIDIA H200 | 4bit | 141 GB | 8.9 GB | Yes |
| Apple M1 Ultra | 4bit | — | 8.9 GB | Yes |
| Apple M3 Max | 4bit | — | 8.9 GB | Yes |
| Apple M4 Max | 4bit | — | 8.9 GB | Yes |
| NVIDIA DGX Spark | 4bit | 128 GB | 8.9 GB | Yes |
| Apple M2 Max | 4bit | — | 8.9 GB | Yes |
| NVIDIA A100 80GB | 4bit | 80 GB | 8.9 GB | Yes |
| NVIDIA H100 SXM | 4bit | 80 GB | 8.9 GB | Yes |
| Apple M1 Max | 4bit | — | 8.9 GB | Yes |
| Apple M4 Pro | 4bit | — | 8.9 GB | Yes |
| Apple M3 Pro | 4bit | — | 8.9 GB | Yes |
| Apple M1 Pro | 4bit | — | 8.9 GB | Yes |
| Apple M2 Pro | 4bit | — | 8.9 GB | Yes |
| Apple M4 | 4bit | — | 8.9 GB | Yes |
| NVIDIA GeForce RTX 5090 | 4bit | 32 GB | 8.9 GB | Yes |
| Apple M2 | 4bit | — | 8.9 GB | Yes |
| Apple M3 | 4bit | — | 8.9 GB | Yes |
| NVIDIA GeForce RTX 3090 | 4bit | 24 GB | 8.9 GB | Yes |
| NVIDIA GeForce RTX 4090 | 4bit | 24 GB | 8.9 GB | Yes |
| Apple M1 | 4bit | — | 8.9 GB | Yes |
No lineage recorded
Papers 1
- arXiv:2412.08905Active35
Repositories 0
No repositories linked yet.
As of
Rewind the record: see this entity's attributes exactly as AI Atlas knew them on a given day.
Claim history · Context window
Context windowcontext_length1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| 16.4K tokens | → current | current | OpenRouter public model & pricing listingT2 | medium | deterministic |
Claims are temporal and append-only: a new observation closes the previous claim (valid_to) instead of overwriting it. Conflicting claims from different sources are kept side by side and flagged — never averaged. Methodology →
Phi 4: release date changed from 2025-01-10 to 2024-12-11
Release date10 Jan 2025→11 Dec 2024huggingfaceOpenRouter lists Phi 4 at $0.07 in / $0.14 out per 1M tokens
openrouter
| Source | Document | Type | Tier | Last observed | Snapshots |
|---|---|---|---|---|---|
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/microsoft/phi-4 | model_page | T2· Quality secondary | 32 min ago | 2 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/models?author=microsoft&p=0&sort=downloads | listing | T2· Quality secondary | 1 h ago | 2 |
| OpenRouter public model & pricing listing | openrouter.ai/api/v1/models | catalogue | T2· Quality secondary | 1 h ago | 1 |
Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.