Microsoft Phi
Hacker ApprovedRanked 5 of 11 small language models by recommenders
Small, MIT-licensed Microsoft models good at simple tasks, math, logic and local tool calling; critics say they are weak elsewhere and dated.
Small, MIT-licensed Microsoft models good at simple tasks, math, logic and local tool calling; critics say they are weak elsewhere and dated.
23 people gave an opinion
- recommend it
- 16
- criticize it
- 7
- approval share
- 70%
Recommended in 11 of 13 months
- November 2025: recommended
- December 2025: recommended
- January 2026: recommended
- February 2026: recommended
- March 2026: recommended
- April 2026: not recommended
- May 2026: recommended
- June 2026: recommended
- July 2026: recommended
- August 2026: recommended
- September 2026: recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it carries the badge
- 16 people recommend itNeeds 10 or more, met.
- 2.3 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 11 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Runs locally in small memory
“I've done pretty decent local prose->json extraction using Qwen and Phi and Gemma.”
Comment from 28 Aug 2026 in “Small Models Have Arrived” “fits in about 3.5gb of RAM (not vram)”
Comment from 3 Jun 2026 in “Gemma 4 12B: A unified, encoder-free multimodal model” “I already use Microsoft Phi offline”
Comment from 2 Jun 2026 in “Microsoft Announces AI Autopilot” Good for simple tasks
“microsofts phi models are very good smaller models under MIT license”
Comment from 7 Nov 2025 in “Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model” “IMO models like Phi-4 are underrated for many simple tasks”
Comment from 20 Dec 2025 in “Big GPUs don't need big PCs” “the small model punching above its weight via data quality”
Comment from 24 Feb 2026 in “Show HN: AI Timeline – 171 LLMs from Transformer (2017) to GPT-5.3 (2026)” Curated training data gives strong results
“Phi was smart (or had a smart training architecture, more precisely), and arguably pushed the conversation forward globally on the value of curated training data.”
Comment from 27 Jul 2026 in “MAI-Cyber 1” “the small model punching above its weight via data quality”
Comment from 24 Feb 2026 in “Show HN: AI Timeline – 171 LLMs from Transformer (2017) to GPT-5.3 (2026)” Recent versions better, less sloppy
“Gemma and phi have gotten better at it in recent versions”
Comment from 21 Jan 2026 in “Which AI Lies Best? LLMs play a 1950s betrayal game by John Nash” “better, less sloppy, more natural language style”
Comment from 4 Sep 2026 in “How concerned should we be about Astra's recurrent architecture?”
What people warn about
Each mark is one comment
Feels dated, outclassed by newer models
“they are much better than things like the Phi series were”
Comment from 7 Jan 2026 in “A 30B Qwen Model Walks into a Raspberry Pi and Runs in Real Time” “Phi was cool for what it was. But it's not 2024 anymore.”
Comment from 27 Jul 2026 in “MAI-Cyber 1” “Microsoft Phi 4 hasn't seen an update in a year.”
Comment from 15 Dec 2025 in “Microsoft AI” Loses originality, restricted outputs
“That's really good to control if you go for strict limits but it also loses a lot of originality.”
Comment from 8 Oct 2026 in “Meta and Microsoft take steps to reduce employee usage of Claude AI” “a lobomised model that is probably Phi-4 :(”
Comment from 29 Oct 2026 in “Gemini 4 Carbon”
Other small language models people recommend
The four most recommended other small language models, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| Claude Haiku | 218 | 134 | 62% | 13 | 4 mo4 owners | — |
| OpenAI Luna | 218 | 56 | 80% | 7 | — | Approved |
| Apple Foundation Models | 45 | 35 | 56% | 11 | 1 yr4 owners | — |
| Ministral | 18 | 6 | 75% | 8 | — | Approved |
| Microsoft PhiThis page | 16 | 7 | 70% | 11 | — | Approved |
What people said
24 of the 27 comments that mention it, newest first. Each links to the original comment.
“a lobomised model that is probably Phi-4 :(”
“IIRC Phi has a tiny context windows.”
“That's really good to control if you go for strict limits but it also loses a lot of originality.”
“Yeah, they have Phi but offer it nowhere on CoPilot as far as I know”
“They're pretty good at searching my personal files”
“I guess MAI models (Phi?) aren't good enough?”
“better, less sloppy, more natural language style”
“I've done pretty decent local prose->json extraction using Qwen and Phi and Gemma.”
“currently does a good job as it is”
“Phi was pretty revolutionary for its time.”
“Phi was smart (or had a smart training architecture, more precisely), and arguably pushed the conversation forward globally on the value of curated training data.”
“Phi was cool for what it was. But it's not 2024 anymore.”
“games all the benchmarks, not great in practice”
“fits in about 3.5gb of RAM (not vram)”
“I already use Microsoft Phi offline”
“Phi 4 mini is good for math and logic, and the reasoning version is surprisingly good at tool calling/RAG, but the family sucks at everything else”
“Phi-4-14b with layers duplicated (phi-4-25b) has increassed performance.”
“the small model punching above its weight via data quality”
“Gemma and phi have gotten better at it in recent versions”
“they are much better than things like the Phi series were”
“IMO models like Phi-4 are underrated for many simple tasks”
“Microsoft Phi 4 hasn't seen an update in a year.”
“they basically didn't follow the rubric very well”
“microsofts phi models are very good smaller models under MIT license”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.