The best small language models, according to hackers
11 small language models were recommended on Hacker News between November 2025 and November 2026. We rank them by how many separate people recommend each one, and 3 earn the Hacker Approved badge.
Top picks
The three most recommended small language models with the badge. Each figure is 25 people. RecommendsCriticizes One comment
Extremely cheap, great value
A very cheap OpenAI model that people say handles everyday and agentic tasks nearly as well as bigger models; critics say it struggles with complex reasoning and coding.
Fast responses for small tasks
Small free models people run locally for fast data extraction and simple tasks; critics say they trail larger models and are weak at vision.
Runs locally in small memory
Small, MIT-licensed Microsoft models good at simple tasks, math, logic and local tool calling; critics say they are weak elsewhere and dated.
All 11 small language models, ranked
Select a column heading to sort. Approval share is recommenders out of everyone who gave an opinion; the tick on each bar marks the 2 to 1 line the badge needs. Months counts the months, out of 13, in which someone recommended it.
| Rank | |||||||
|---|---|---|---|---|---|---|---|
| Tied 1 | Claude Haiku | 218 | 134 | 62% | 13 | 4 mo4 owners | — |
| Tied 1 | OpenAI Luna | 218 | 56 | 80% | 7 | — | Approved |
| 3 | Apple Foundation Models | 45 | 35 | 56% | 11 | 1 yr4 owners | — |
| 4 | Ministral | 18 | 6 | 75% | 8 | — | Approved |
| 5 | Microsoft Phi | 16 | 7 | 70% | 11 | — | Approved |
| 6 | LFM2.5 2.6B | 14 | 8 | 64% | 9 | — | — |
| 7 | Liquid AI 8B-A1B MoE | 12 | 7 | 63% | 5 | — | — |
| 8 | Chrome built-in AI model | 11 | 17 | 39% | 2 | — | — |
| Tied 9 | Gemini Nano | 10 | 8 | 56% | 4 | — | — |
| Tied 9 | LFM3 | 10 | 2 | 83% | 1 | — | — |
| 11 | Desert Ant Labs local models | 7 | 3 | 70% | 1 | — | — |
How we rank
Rank is the number of separate people who recommend a product in a Hacker News comment over the past year. Someone who praises the same small language model in five threads counts once.
The badge goes to small language models that at least 10 people recommend, with recommenders outnumbering critics at least 2 to 1, recommended in at least 4 of the 13 months.
Rankings are never for sale. We may earn a commission when you buy through our links. It never changes the order.