LOADING DATASHEET
GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH
Phi 4
◆ OPEN
by Microsoft
OPEN WEIGHTSLOCALAPIPAIDCLOSED
RANKED #172 OF 271 ASSISTANTS · OVERALL #274 OF 6,567 · VIBE SCORE 5.6 · 101 VOICES
[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...
HAPPY8.2
18 mentions
SAD6.8
8 mentions
VIBE TRAJECTORY
26 weeks · 56 voices
HONEYMOON FADING
Early praise is cooling off in recent weeks.
HOW TO USE IT
Open weights. You can use a hosted service, or download it and run it yourself, free.
RUN IT ON YOUR OWN MACHINE
Open weights: download Phi 4 once and it's yours. No subscription, no rate limits, works offline.
14B parameters, a Q4 file is about 9 GB
- BARE MINIMUM
NVIDIA GeForce RTX 3060 12GB
12 GB VRAM
~$220 USEDQ4_K_M, ~25 tok/s, 16k context
- SWEET SPOT
NVIDIA GeForce RTX 4060 Ti 16GB
16 GB VRAM
~$380 USEDQ8_0, ~35 tok/s, 16k context
- FULL POWER
NVIDIA GeForce RTX 4090 24GB
24 GB VRAM
~$2,250 USEDQ8_0 or FP16 with partial offload, ~60 tok/s, 16k context
LEARN HOW TO SET IT UP
WILL MY BOX RUN PHI 4?
The three cards above are researched picks. Search the machine you actually have — we only say yes if it should reply at a usable speed, not just load the weights and crawl.
HARDWARE PICKS & PRICES ARE RESEARCHED FROM THE LIVE WEB AND REFRESHED AUTOMATICALLY. TREAT THEM AS BALLPARK, NOT GOSPEL.
WHAT CAN YOU ACTUALLY DO WITH IT?
Everyday jobs on each of those rigs: how long Phi 4 takes, and how good the crowd says it is at that kind of work.
| THE JOB | BARE MINIMUMNVIDIA GeForce RTX 3060 12GB | SWEET SPOTNVIDIA GeForce RTX 4060 Ti 16GB | FULL POWERNVIDIA GeForce RTX 4090 24GB |
|---|---|---|---|
| Write an emaila paragraph or two, drafted from a one-line brief | ~20 S★★★★★ | ~14 S★★★★★ | ~8 S★★★★★ |
| Summarize a documenta long report boiled down to the points that matter | ~60 S★★★★★ | ~43 S★★★★★ | ~25 S★★★★★ |
| Build a websitea small landing page, markup and styles together | ~5.3 MIN★★★★★ | ~3.8 MIN★★★★★ | ~2.2 MIN★★★★★ |
| Build a backendan API with routes, storage and tests | ~17 MIN★★★★★ | ~12 MIN★★★★★ | ~6.9 MIN★★★★★ |
| Build a gamea playable browser game in one file | ~10 MIN★★★★★ | ~7.1 MIN★★★★★ | ~4.2 MIN★★★★★ |
STARS ARE THE COMMUNITY SCORE FOR THAT KIND OF WORK, DOCKED FOR HOW SQUASHED THE WEIGHTS GET AT EACH TIER. TIMES ARE COMPUTED FROM RESEARCHED SPEEDS. BALLPARK, NOT GOSPEL.
WHAT THE INTERNET SAYS
Not enough discussion about Phi 4 yet to call it. We found 101 posts but people did not say much either way.
74 POSTS · 27 COMMENTS · REDDIT · OTHER FORUMS · HACKER NEWS · GITHUB · X · LEMMY
PEOPLE LOVE
- No fan mail yet. The crowd is still warming up.
PEOPLE COMPLAIN ABOUT
- No complaints yet. The internet is still looking for one. Suspicious.
LOOKING FOR AN ALTERNATIVE?
Checked picks first: tags say why you would switch, stars say how fully each one stands in for Phi 4.
Nemotron 3.5 Lightning
LFM2.5 2.6B (free)
Grok 4.6> SOURCES (12)
- REDDIT · 2025-02-07 · 773 UPVOTESYou can now train your own o3-mini model on your local device!
You can now train your own o3-mini model on your local device!
- REDDIT · 2025-01-13 · 361 UPVOTESI fixed 4 bugs in Microsoft's open-source Phi-4 model
I fixed 4 bugs in Microsoft's open-source Phi-4 model
- REDDIT · 2025-01-15 · 269 UPVOTES[P] How I found & fixed 4 bugs in Microsoft's Phi-4 model
[P] How I found & fixed 4 bugs in Microsoft's Phi-4 model
- REDDIT · 2025-06-06 · 207 UPVOTES[R] LLMs are Locally Linear Mappings: Qwen 3, Gemma 3 and Llama 3 can be converted to exactly equivalent locally linear systems for…
[R] LLMs are Locally Linear Mappings: Qwen 3, Gemma 3 and Llama 3 can be converted to exactly equivalent locally linear systems for interpretability
- REDDIT · 2025-02-08 · 196 UPVOTESTrain your own Reasoning model like DeepSeek-R1 locally (7GB VRAM min.)
Train your own Reasoning model like DeepSeek-R1 locally (7GB VRAM min.)
- REDDIT · 2025-02-07 · 188 UPVOTESYou can now train your own DeepSeek-R1 model on your local device!
You can now train your own DeepSeek-R1 model on your local device!
- REDDIT · 2026-08-08 · 117 UPVOTESIs Microsoft-Phi dead?
Is Microsoft-Phi dead?
- REDDIT · 2025-05-01 · 87 UPVOTESYou can now run 'Phi-4 Reasoning' models on your own local device! (20GB RAM min.)
You can now run 'Phi-4 Reasoning' models on your own local device! (20GB RAM min.)
- REDDIT · 2025-01-18 · 81 UPVOTESPHI-4: Quantized Q4 vs Q8 on My Nvidia RTX 3060 12GB System
PHI-4: Quantized Q4 vs Q8 on My Nvidia RTX 3060 12GB System
- REDDIT · 2025-02-20 · 78 UPVOTESTrain your own Reasoning model like DeepSeek-R1 locally (5GB VRAM min.)
Train your own Reasoning model like DeepSeek-R1 locally (5GB VRAM min.)
- REDDIT · 2026-04-20 · 72 UPVOTESI benchmarked 21 local LLMs on a MacBook Air M5 for code quality AND speed
I benchmarked 21 local LLMs on a MacBook Air M5 for code quality AND speed
- HACKER NEWS · 2026-04-29 · 60 UPVOTESShow HN: A new benchmark for testing LLMs for deterministic outputs
Show HN: A new benchmark for testing LLMs for deterministic outputs