GoogleBlog · Jul 8, 2026 · official
Performance of medium-sized LLMs
Performance of medium-sized LLMs (post-training) from 13B to 100B
BlogGrouped bar chartGemma specialist modelsAll Google modelsTakumi-8B-gpt-4oLlama-3-Swallow-70B-instructLlama-3-70B-instructCyberAgentLM3-3-9B-1.5-instructGPT-3.5-Turbo-0125Llama-3.1-Swallow-70B-instructLlama-3-70B-instruct-v0.1Aya-Expanse-32B
API
Not in the API yet
We have not read this chart's numbers yet, so the API has nothing to return for it. Charts with numbers read are on the data page, and each one can be called by id.
More charts about Gemma specialist models
More from Google
3.2xGoogle·Oct 20252.9M views14.6K likes0.6% eng.
OfficialX · GoogleFeb 12, 2026
Gemini 3 Deep Think on Humanity's Last Exam695.4K views · 1.9K likes
Questions people ask
Where does this chart come from?
Google published it on Jul 8, 2026 on Blog (deepmind.google). We keep it here, with the publisher and date credited as text, so you can study it without leaving.
Can I make a chart like this?
Use the format, not the picture. The maker on any chart page redraws this type with your own numbers and words.