whatmodel.sh / Flan-T5 XL

Flan-T5 XL

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Flan-T5 XL is an open-weights language model from Google, released October 2022. Its published size is 2.85B parameters. Licence: apache-2.0.

Catalogue identity notes

Instruction-finetuned T5 v1.1 LM-adapted checkpoint released with Chung et al. 2022.

31.1±1.5#633 of 719 LLMsfrom 5 benchmark results across 2 sources

$ how it compares

The LLMs ranked either side of Flan-T5 XL, out of 828 with a score, then more from Google.

#modelscorewhereRAMreleased
728Phi-1.5 (1.3B) Microsoft32±1local~2.4GB2023-09
729Falcon-H1 1.5B Instruct TII31±4local~2.4GB2025-05
730Flan-T5 XL Google31±2local~14GB2022-10
731Qwen2 1.5B Instruct Alibaba31±1local~2.4GB2024-06
732Qwen-7B-Chat Alibaba31±1local~6.6GB2023-08
more from Google
9Gemini 4 Argon (limited preview) Google82±1limited preview2026-09
23Gemini 3.8 Flash Google77±1api $1.50/M2026-09
32Gemini 3.7 Flash Google76±1api $1.50/M2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

math██·█···█···█ #2 of 5 * · 15% of its score · 1 test
testresultvs the others in its rundifficultycountswhy
MGSM lab1.9% ✎▃·▃·▁··▂▄█▄▄ beats 3/26hard · avg 22%▇▇▇▇▇▇ 15%too hard for it
other tests2 not in a task · 84% of its score
testresultvs the others in its rundifficultycountswhy
BIG-Bench Hard +140.2% ✎███··█·█·█·█ beats 5/6easy · avg 80%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 56%sharp · several runs
MMLU +149.2% ✎▄▄·▄···▄▄·██ beats 7/8easy · avg 82%▇▇▇▇▇▇▇▇▇▇▇ 28%at its level · several runs

$ where to run it

Other ways to run it

  1. Run it as published · needs ~11.4 GB · from huggingface.co/google (official)
    uv run --with torch --with 'transformers<5' --with accelerate python -c "from transformers import pipeline; print(pipeline('text2text-generation', model='google/flan-t5-xl', torch_dtype='auto', device_map='auto')('Answer the question: what is the capital of France?', max_new_tokens=40)[0]['generated_text'])"

Official weights and documentation: huggingface.co/google ↗

$ how the score was worked out · Scaling Instruction-Finetuned… 66%, InstructEval (DeCLaRe Lab) RE… 33%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
Scaling Instruction-Finetuned Language Models — authors' own runs · lab-reported3▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 66%CC BY 4.0
InstructEval (DeCLaRe Lab) README results table2▇▇▇▇▇▇▇▇▇▇ 33%MIT / Apache-2.0
prior · pull toward the middle▇ 1%
weighted average= 31.1

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json