whatmodel.sh / GLaM 1.7B/64E (MoE)

GLaM 1.7B/64E (MoE)

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

GLaM 1.7B/64E (MoE) is a closed-weights language model from Google, released December 2021. Its published size is 27B parameters (1.879B active per token). Licence: proprietary.

Catalogue identity notes

Google's GLaM 1.7B/64E mixture-of-experts model (Du et al., arXiv:2112.06905, Table 4: 27B total / 1.879B activated parameters), trained on the GLaM dataset. Research size from the same paper. Never released. Not the same checkpoint as LaMDA-PT even where sizes match. Comparison-only.

Model information sources

Release date ↗

53.0±1.3comparison only: no current run option verifiedfrom 29 benchmark results across 1 source

Score within the ANLI Round 1 (prompted) comparison set only.

$ how it compares

Scored LLMs released closest to GLaM 1.7B/64E (MoE), out of 1182 with a score, then more from Google.

#modelscorewhereRAMreleased
1034LaMDA 137B (pre-trained, LaMDA-PT) Google34±1never released2022-01
1035GPT-3 text-davinci-001 OpenAI34±1comparison only2022-01
1106LaMDA 68B (pre-trained) Google27±1never released2022-01
1156LaMDA 8B (pre-trained) Google12±2never released2022-01
1158GPT-3 text-curie-001 OpenAI12±1comparison only2022-01
more from Google
11Gemini 4 Argon Google80±1api $4.00/M2026-09
26Gemini 3.8 Flash Google76±1api $1.50/M2026-09
34Gemini 3.7 Flash Google75±1api $1.50/M2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

other tests29 not in a task · 99% of its score
testresultvs the others in its rundifficultycountswhy
TriviaQA lab44.0% ✎▄··▄···▄·█▄█ beats 3/7hard · avg 35%▇▇▇▇▇▇ 15%sharp · at its level
Natural Questions (closed book) lab9.2% ✎▄·▄··▄·█▄·▄▄ beats 4/7very hard · avg 7.2%▇▇▇ 8.3%sharp
ARC Challenge lab37.6% ✎▂··▂·····▄·█ beats 3/7very hard · avg 35%▇▇▇ 7.3%sharp
WinoGrande lab63.9% ✎▃··▃·····▃▅█ beats 3/7hard · avg 61%▇▇▇ 7.2%sharp · at its level
HellaSwag lab65.8% ✎▃··▃···▃·▃▃█ beats 3/7medium · avg 62%▇▇▇ 7.1%sharp · at its level
SQuAD 2.0 (prompted, F1) lab49.2% ✎██····██████ beats 3/7medium · avg 47%▇▇▇ 6.8%sharp · at its level
WebQuestions (closed book) lab8.3% ✎▄····▄·▄█▄▄▄ beats 4/7very hard · avg 6.7%▇▇ 5.3%sharp · too hard for it
COPA lab85.0% ✎▄·▄···▄·▄█·█ beats 3/7easy · avg 83%▇▇ 4.2%at its level
21 more tests
testresultvs the others in its rundifficultycountswhy
RACE-high lab40.4% ✎▃▃······▅·▃█ beats 3/7hard · avg 39%▇▇ 3.8%typical
Winograd Schema Challenge (WSC273) lab80.2% ✎█·····▄▄▄▄▄▄ beats 3/7medium · avg 79%▇ 3.5%at its level
BoolQ lab62.7% ✎█···▃··▃▃·▃▃ beats 3/7hard · avg 65%▇ 3.4%sharp · at its level · noisy
SuperGLUE WSC (prompted) lab79.3% ✎█·····▄·▄█▄▄ beats 3/7medium · avg 77%▇ 3.2%at its level
Story Cloze Test lab76.2% ✎▄·▄····▄▄▄▄█ beats 3/7medium · avg 75%▇ 3.2%at its level
RACE-middle lab56.1% ✎▄▄····▄▄·▄█▄ beats 3/7medium · avg 56%▇ 2.9%at its level
ARC Easy lab66.2% ✎▄··▄···▄·█▄█ beats 4/7medium · avg 63%▇ 2.6%at its level
CoQA lab65.3% ✎▃▃·····█▃▃·▃ beats 2/7medium · avg 64%▇ 2.4%at its level
ReCoRD (SuperGLUE, prompted) lab87.1% ✎▄·▄·····▄▄██ beats 3/7very easy · avg 86%▇ 2.0%typical
DROP lab13.4% ✎▄·▄██▄·····▄ beats 2/7hard · avg 16%▇ 1.9%sharp · noisy
LAMBADA (OpenAI version), last-word accuracy lab63.7% ✎▄▄·····▄·█▄█ beats 3/7medium · avg 60%▇ 1.9%at its level
PIQA lab76.9% ✎▃···▃··▃·▃█▃ beats 3/7medium · avg 75%▇ 1.9%at its level
OpenBookQA lab46.4% ✎▄·▄····█·█·█ beats 2/7hard · avg 46%▇ 1.8%at its level
RTE (prompted) lab58.8% ✎█▄··▄····▄▄█ beats 4/7very hard · avg 55%▇ 1.3%sharp · noisy
QuAC (prompted, F1) lab32.8% ✎█···▄·█▄··▄▄ beats 3/7hard · avg 32%▇ 0.8%blunt · at its level
ANLI Round 3 (prompted) lab34.0% ✎▄·▄▄██·····▄ beats 1/7very hard · avg 35%▇ 0.7%sharp · too hard for it · noisy
ANLI Round 1 (prompted) lab33.9% ✎█·▄▄·▄··▄··█ beats 2/7very hard · avg 35%▇ 0.3%too hard for it · noisy
MultiRC (SuperGLUE, prompted, F1a) lab58.0% ✎▃···▅▃█····▃ beats 5/7medium · avg 56%▇ 0.1%blunt · at its level · noisy
ANLI Round 2 (prompted) lab32.4% ✎█▄·····█·▄·█ beats 1/7very hard · avg 35%▇ 0.1%blunt · too hard for it · noisy
CommitmentBank (prompted) lab33.9% ✎█▄▄█·▄·····▄ beats 1/7medium · avg 43%▇ 0.0%blunt · at its level · noisy
WiC (prompted) lab49.8% ✎█▄·······▂·▂ beats 2/7very hard · avg 50%▇ 0.0%blunt · too hard for it · noisy

$ where to run it

No verified API or download recorded yet · add one

$ how the score was worked out · GLaM: Efficient Scaling of La… 99%

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
GLaM: Efficient Scaling of Language Models with Mixture-of-Experts — authors' own runs · lab-reported29CC BY 4.0
prior · pull toward the middle
weighted average= 53.0

$ 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