- Size
- 14.7B parameterscounted from the safetensors header on the Hub
- Licence
- custom, read it, MIT LicenseOur reading, check the licence: A custom licence we have not read for this model: other makers use the same name for different texts. Read the model's own licence file before you use it.Licence text
- On the Hub since
- 11 Dec 2024the date the repo was created
- Context length
- 16k tokensfrom the model's config
- Kind
- Base model (no base model declared on the Hub)Phi3ForCausalLM
- On Hugging Face
- microsoft/phi-4192k downloads in the last 30 days, 2.3k likes
Scores
- default setting56.1%±2.631 Jan 2025, log
Epoch's name: phi-4. Best closed model on this set: gpt-6.2-astra_max, 96.8% (OpenAI). All 341 results
- default setting64.9%±1.131 Jan 2025, log
Epoch's name: phi-4. Best closed model on this set: gpt-5-2025-08-07_high, 98.1% (OpenAI). All 108 results
- default setting13.8%±3.725 Feb 2025, log
Epoch's name: phi-4. Best closed model on this set: claude-fable-5_high, 100.0% (Anthropic). All 318 results
- default setting11.6%±1.413 Jan 2025, balrogai.com
Epoch's name: phi-4. Best closed model on this set: gpt-6-astra_max, 68.3% (OpenAI). All 41 results
- default setting0.0%via Epoch AI
Epoch's name: phi-4. Best closed model on this set: gemini-4-carbon_high, 36.0% (Google DeepMind). All 271 results
- default setting62.6%lechmazur/writing Github repository
Epoch's name: phi-4. Best closed model on this set: gpt-5-2025-08-07_medium, 86.0% (OpenAI). All 49 results
- default setting41.6%LiveBench Leaderboard
Epoch's name: phi-4. Best closed model on this set: gemini-2.5-pro-exp-03-25, 82.3% (Google DeepMind). All 54 results
- Shots: 584.8%Phi-4 Technical Report
Epoch's name: phi-4. Best closed model on this set: gpt-4o-2024-11-20, 88.1% (OpenAI). All 217 results
Claimed by the maker Claimed
The model card has no claimed scores in a machine-readable form (the card's model-index), so none are shown. The maker may give scores in the card text: read the card.
Other views, which we link to and do not copy: look this model up on Artificial Analysis and the Arena leaderboard.
Independent scores from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Each row links its own source. What the badges mean.
Run it yourself
Memory needed = weights + context cache + 0.50 GB for the runtime. Context cache at 8k tokens: 1.6 GB, worked out from the model's config (microsoft/phi-4), at 16-bit precision. Choose a machine above to see a verdict for each format. Method
GGUF for llama.cpp, Ollama, LM Studio and others
MaziyarPanahi/phi-4-GGUF (converted by MaziyarPanahi), 40k downloads. Sizes are the real files on the Hub, all parts added up.
| Format | Size | Memory needed | Get it |
|---|---|---|---|
| Q2_K | 5.2 GBfile size | 7.2 GB5.2 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:Q2_KOllama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q2_KDownload hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.Q2_K.gguf" --local-dir phi-4-GGUF |
| Q3_K_S | 6.1 GBfile size | 8.1 GB6.1 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:Q3_K_SOllama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q3_K_SDownload hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.Q3_K_S.gguf" --local-dir phi-4-GGUF |
| Q3_K_M | 6.9 GBfile size | 8.9 GB6.9 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:Q3_K_MOllama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q3_K_MDownload hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.Q3_K_M.gguf" --local-dir phi-4-GGUF |
| Q3_K_L | 7.4 GBfile size | 9.4 GB7.4 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:Q3_K_LOllama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q3_K_LDownload hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.Q3_K_L.gguf" --local-dir phi-4-GGUF |
| Q4_K_S | 7.9 GBfile size | 9.9 GB7.9 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:Q4_K_SOllama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q4_K_SDownload hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.Q4_K_S.gguf" --local-dir phi-4-GGUF |
| Q4_K_M | 8.4 GBfile size | 10.5 GB8.4 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:Q4_K_MOllama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q4_K_MDownload hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.Q4_K_M.gguf" --local-dir phi-4-GGUF |
| Q5_K_S | 9.5 GBfile size | 11.5 GB9.5 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:Q5_K_SOllama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q5_K_SDownload hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.Q5_K_S.gguf" --local-dir phi-4-GGUF |
| Q5_K_M | 9.9 GBfile size | 11.9 GB9.9 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:Q5_K_MOllama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q5_K_MDownload hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.Q5_K_M.gguf" --local-dir phi-4-GGUF |
| Q6_K | 11.2 GBfile size | 13.3 GB11.2 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:Q6_KOllama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q6_KDownload hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.Q6_K.gguf" --local-dir phi-4-GGUF |
| Q8_0 | 14.5 GBfile size | 16.6 GB14.5 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:Q8_0Ollama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q8_0Download hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.Q8_0.gguf" --local-dir phi-4-GGUF |
| F16 | 27.3 GBfile size | 29.4 GB27.3 + 1.6 + 0.50 | Commandsllama.cpp llama-server -hf MaziyarPanahi/phi-4-GGUF:F16Ollama ollama run hf.co/MaziyarPanahi/phi-4-GGUF:F16Download hf download MaziyarPanahi/phi-4-GGUF --include "phi-4.fp16.gguf" --local-dir phi-4-GGUF |
MLX for Apple silicon
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| MLX 4bitmlx-community/phi-4-4bit | 7.7 GBfile size | 9.7 GB7.7 + 1.6 + 0.50 | Download page |
| MLX 8bitmlx-community/phi-4-8bit | 14.5 GBfile size | 16.6 GB14.5 + 1.6 + 0.50 | Download page |
Original weights size from the safetensors header, for vLLM, SGLang, Transformers
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| BF16original weights | 27.3 GBfile size | 29.4 GB27.3 + 1.6 + 0.50 |
hf download microsoft/phi-4 --local-dir phi-4Other conversions (14; sizes not read)
- RedHatAI/phi-4-FP8-dynamic FP8, 14k downloads
- unsloth/phi-4-unsloth-bnb-4bit bitsandbytes, 4bit, 3.0k downloads
- unsloth/phi-4-bnb-4bit bitsandbytes, 4bit, 2.9k downloads
- Tomhn/phi-4-FP8-Dynamic-fixed FP8, 2.5k downloads
- stelterlab/phi-4-AWQ AWQ, 1.6k downloads
- RedHatAI/phi-4-quantized.w8a8 Other, w8a8, 915 downloads
- RedHatAI/phi-4-quantized.w4a16 Other, w4a16, 898 downloads
- jkim96/phi-4-DASHQ-INT4-g128 Other, int4, 146 downloads
- jkim96/phi-4-DASHQ-INT3-g128 Other, 143 downloads
- Satwik11/Microsoft-phi-4-Instruct-AutoRound-GPTQ-4bit GPTQ, 4bit, 82 downloads
- OPEA/phi-4-int4-AutoRound-gptq-sym GPTQ, int4, 76 downloads
- curiousmind147/microsoft-phi-4-AWQ-4bit-GEMM AWQ, 4bit, 75 downloads
- aleada/Phi-4-W4A16 GPTQ, w4a16, 64 downloads
- cortecs/phi-4-FP8-Dynamic FP8, 45 downloads
The Hub tags 100 or more quantised repos as conversions of this model; we read the most downloaded and show the maker's own first.
Run it by API
2 providers serve this model through the Hugging Face router. Prices are US dollars per million tokens. Precision is what the provider reports; a lower precision than the original weights can change answers. Live first, then cheapest output.
| Provider | In | Out | Context | Precision | Tokens/s | First token | Uptime 24h | Tools | JSON |
|---|---|---|---|---|---|---|---|---|---|
| DeepInfra | $0.07 | $0.14 | 16k | bf16 | 94 | 0.3 s | 100.0% | no | yes |
| featherless-ai | ? | ? | ? | not given | ? | ? | ? | not given | not given |
Through the router, with a Hugging Face token (the router bills your Hugging Face account). Change the name after the colon to pick another provider.
curl https://router.huggingface.co/v1/chat/completions \
-H "Authorization: Bearer $HF_TOKEN" -H "Content-Type: application/json" \
-d '{"model": "microsoft/phi-4:deepinfra", "messages": [{"role": "user", "content": "Hello"}]}'Not built yet: one key for every open model, with BlockRun, and our own tested-quality score per provider. Until then, precision and measured speed are the best guide to a provider here.
Fine-tunes of this model
Repos that declare this model as their base, most downloaded first. Listing is not a recommendation; unreviewed ones have not been checked.
- microsoft/Phi-4-reasoning14.7B33kUnreviewed
- microsoft/Phi-4-reasoning-plus14.7B4.6kUnreviewed
Planned for this page Not built yet
- Try it in private chat
- A link to Solana AI's confidential chat for the models it serves. This model is not listed as served there.
- Fine-tune it
- Fine-tuning a few base models from this page and publishing the result here.
- Creators and earnings
- Pages for people who publish fine-tunes and adapters of this model, and what they earn.
- Fund and bounties
- Bounties for new open models, paid from a public fund. How the fund will work.
- Contribute compute
- How to give compute to open model work, with Akash.
Model data from the Hugging Face Hub 19 Nov 2026, 13:43 UTC. The repo. Found a mistake on this page? Tell us.