Title: Phi 3 Vision 128K Instruct vs Phi-4: Parametri, Licen%C8%9B%C4%83 %E2%80%94 LLM EU
Canonical: https://llmeu.com/ro/compare/phi-3-vision-128k-instruct-vs-phi-4/
Markdown mirror: https://llmeu.com/ro/compare/phi-3-vision-128k-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 3 Vision 128K Instruct vs Phi-4

Pune%C8%9Bi dou%C4%83 p%C3%A2n%C4%83 la patru modele al%C4%83turi pe c%C3%A2mpurile care decid o implementare: context, pre%C8%9B, regiune, licen%C8%9B%C4%83, limbi, g%C4%83zduire %C8%99i ce este verificat.

Acolo unde dou%C4%83 modele difer%C4%83, acest tabel arat%C4%83 diferen%C8%9Ba %C8%99i se opre%C8%99te. LLM EU nu public%C4%83 nicio insign%C4%83 de c%C3%A2%C8%99tig%C4%83tor f%C4%83r%C4%83 un punctaj documentat pentru acea sarcin%C4%83, iar o cifr%C4%83 lips%C4%83 r%C4%83m%C3%A2ne lips%C4%83 %C3%AEn loc s%C4%83 devin%C4%83 zero.

## C%C3%A2nd se folose%C8%99te

### [Phi 3 Vision 128K Instruct](/ro/models/phi-3-vision-128k-instruct/)

#### Puncte forte

- 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 4.1B parameters it fits on a single accelerator
- instruction-tuned checkpoint with a chat template in the repository

#### Limit%C4%83ri

- 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 4.1B parameters, the weights fit on a single modest accelerator

#### Sarcini

[Viziune](/ro/tasks/vision/), [Reg%C4%83sire (RAG)](/ro/tasks/rag/)

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

#### Puncte forte

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

#### Limit%C4%83ri

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

#### Sarcini

[Ra%C8%9Bionament](/ro/tasks/reasoning/), [Sumarizare](/ro/tasks/summarization/), [Cost redus](/ro/tasks/cheap/), [Reg%C4%83sire (RAG)](/ro/tasks/rag/)

**Sarcini:** [Reg%C4%83sire (RAG)](/ro/tasks/rag/)

## Compara%C8%9Bi modele

- [Aya 23 8B vs Phi-4](/ro/compare/aya-23-8b-vs-phi-4/)
- [Aya Expanse 8B vs Phi-4](/ro/compare/aya-expanse-8b-vs-phi-4/)
- [Blip2 Flan T5 Xl vs Phi 3 Vision 128K Instruct](/ro/compare/blip2-flan-t5-xl-vs-phi-3-vision-128k-instruct/)
- [Blip2 Opt 2.7B vs Phi 3 Vision 128K Instruct](/ro/compare/blip2-opt-2-7b-vs-phi-3-vision-128k-instruct/)
- [NVIDIA Cosmos 3 Edge vs Phi 3 Vision 128K Instruct](/ro/compare/cosmos-3-edge-vs-phi-3-vision-128k-instruct/)
- [NVIDIA Cosmos Reason2 2B vs Phi 3 Vision 128K Instruct](/ro/compare/cosmos-reason2-2b-vs-phi-3-vision-128k-instruct/)