Models & assistants Model or AI service
EXAONE 3.5 offers instruction-tuned models in three sizes
LG AI Research released EXAONE 3.5 in 2.4B, 7.8B and 32B sizes. Its models are tuned to follow instructions and can be downloaded for research under LG’s terms. [1][2]
- Model access
- Three instruction-tuned sizes are downloadable for research under LG AI Research’s terms. [1][2]
- License
- EXAONE AI Model License Agreement 1.1 - NC for the cited 7.8B checkpoint. [2]
What the model does
LG submitted a report on the three models in December 2024. It described a smaller model for lighter hardware and two larger options for work that needs more capacity. [1][2]
Why it matters
A selection of sizes gives researchers a way to compare the model on the same task under different compute constraints. The middle checkpoint can be evaluated without committing to the 32B option. [1][2]
Who can use it
Researchers can evaluate the Korean- and English-capable models for following instructions or answering questions about longer documents. [1][2]
Key dates
- EXAONE 3.5 technical paper first submitted [1]
Ways to use it
Compare how two sizes answer a question whose evidence appears far into a document. Check whether each response uses the correct passage and measure memory use in the same setup. [1][2]
Model limitations
LG notes that model responses can be false, contradictory or biased. The 7.8B checkpoint has a research-oriented noncommercial license, so a production plan needs a review of the applicable terms. [1][2]
Sources and verification
Catalog evidence checked How we verify records
- Primary source LG AI Research’s first EXAONE 3.5 report and model lineup
- Primary source LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct
- Other source Open-sourcing Three EXAONE 3.5 Models : Frontier-level Model, Top-tier Performance in Instruction Following and Long Con
- Other source EXAONE 3.5: Series of Large Language Models for Real-world Use Cases
Related profiles
Models & assistants
GPT-6.2 Astra and Sol
OpenAI
Astra targets demanding projects. Sol targets everyday tasks. Both support text and image input in the API.
Models & assistants
MAI Code 2 and Code 2 Flash
Microsoft AI
Code 2 targets complex software work. Flash targets frequent edits, explanations and routine fixes.
Models & assistants
MAI Realtime 1
Microsoft AI
Users can try a MAI Labs demo. Developers can register for a limited Foundry preview.