Title: Phi 4 Mini Instruct vs Phi-4: Parametrar, Licens %E2%80%94 LLM EU
Canonical: https://llmeu.com/sv/compare/phi-4-mini-instruct-vs-phi-4/
Markdown mirror: https://llmeu.com/sv/compare/phi-4-mini-instruct-vs-phi-4.md
Machine-readable: https://llmeu.com/llms.txt (index) - https://llmeu.com/llms-full.txt (full catalogue) - https://llmeu.com/llms-catalogue.json

# Phi 4 Mini Instruct vs Phi-4

St%C3%A4ll tv%C3%A5 till fyra modeller sida vid sida p%C3%A5 de f%C3%A4lt som avg%C3%B6r en drifts%C3%A4ttning: kontext, pris, region, licens, spr%C3%A5k, hosting och vad som %C3%A4r verifierat.

D%C3%A4r tv%C3%A5 modeller skiljer sig visar denna tabell skillnaden och stannar. LLM EU publicerar ingen vinnarutm%C3%A4rkelse utan en dokumenterad po%C3%A4ng f%C3%B6r den uppgiften, och en saknad siffra f%C3%B6rblir saknad i st%C3%A4llet f%C3%B6r att bli en nolla.

## N%C3%A4r den ska anv%C3%A4ndas

### [Phi 4 Mini Instruct](/sv/models/phi-4-mini-instruct/)

#### Styrkor

- open weights are published on Hugging Face and can be downloaded without a vendor agreement
- mit licence, so commercial self-hosting is permitted without a separate agreement
- at 3.8B parameters it fits on a single accelerator
- the model page lists 13 languages the European Union publishes in, including cs, da, nl, en
- instruction-tuned checkpoint with a chat template in the repository

#### Begr%C3%A4nsningar

- catalogue only %E2%80%94 LLM EU does not run this model, and nothing on this card implies an LLM EU endpoint
- no evaluation has been run by LLM EU
- the published configuration records no context window, so the card carries none
- at 3.8B parameters, the weights fit on a single modest accelerator

#### Uppgifter

[Resonemang](/sv/tasks/reasoning/), [Sammanfattning](/sv/tasks/summarization/), [Flerspr%C3%A5kigt st%C3%B6d](/sv/tasks/multilingual-support/)

### [Phi-4](/sv/models/phi-4/)

#### Styrkor

- strong MMLU and maths scores at 14B parameters
- MIT licence without user thresholds
- fits a single 40GB accelerator at BF16

#### Begr%C3%A4nsningar

- 16K-token context window
- declared English-only; EU-language performance is not documented
- training leans on synthetic data, which narrows domain robustness

#### Uppgifter

[Resonemang](/sv/tasks/reasoning/), [Sammanfattning](/sv/tasks/summarization/), [L%C3%A5g kostnad](/sv/tasks/cheap/), [H%C3%A4mtning (RAG)](/sv/tasks/rag/)

**Uppgifter:** [Resonemang](/sv/tasks/reasoning/), [Sammanfattning](/sv/tasks/summarization/)

## J%C3%A4mf%C3%B6r modeller

- [Aya 23 8B vs Phi-4](/sv/compare/aya-23-8b-vs-phi-4/)
- [Aya Expanse 8B vs Phi-4](/sv/compare/aya-expanse-8b-vs-phi-4/)
- [Bloom 3B vs Phi 4 Mini Instruct](/sv/compare/bloom-3b-vs-phi-4-mini-instruct/)
- [NVIDIA Cosmos Reason2 2B vs Phi 4 Mini Instruct](/sv/compare/cosmos-reason2-2b-vs-phi-4-mini-instruct/)
- [DeepSeek Coder V2 Lite Instruct vs Phi-4](/sv/compare/deepseek-coder-v2-lite-instruct-vs-phi-4/)
- [DeepSeek R1 0528 Qwen3 8B vs Phi-4](/sv/compare/deepseek-r1-0528-qwen3-8b-vs-phi-4/)