OLMoE-1B-7B-0924

6.92B parameters (8 of 64 experts per token), text in, text out, 4K tokens of context.

Verified 2026-10-25 Rank #744 in LLMs · score 71.6 method v1 Sources 3 Downloads 30d 69.6k · #357 of 619 tracked

Can I run it?4.2GB at INT4Estimatefits a 16 GB cardmethod

Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24

Do people use it?69.6kdownloads, 30 days#357 of 619 we track · below the median+0 likes, downloads −21% this week156 likesHugging Face2026-10-24

Worth reading

Picked by rule from 3 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Latest · 20 months ago

Ai2 OLMoE: an on-device open-source model

Ai2 released a fully open iOS app for the OLMoE model, allowing users to run AI privately on their devices.

The model uses the Dolmino mix for mid-training and the Tülu 3 post-training recipe.
Quantization causes IFEval scores to drop from 66.4 to 63.6 with minimal performance impact.

allenai.org 2025-02-12 5 HN pts · 1 comments 643 words about the OLMoE family family

3 best of 3 tracked

The 1 pick above, plus these 2. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleOLMoE and the hidden simplicity in training better foundation modelsinterconnects.ai · 2024-09-04 · 3 HN pts
ArticleAn open, on-device model powered chat appfor iOSapps.apple.com · 2025-02-11 · 2 HN pts

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

No API provider serves it, so its downloads are people running it themselves: 9 quantized builds and 7 Spaces.

As a service0no API provider listed
On own hardware9quantized builds
Built on117 fine-tunes · 4 adapters · 0 merges
In apps7Hugging Face Spaces

Top Spaces running it: allenai/reward-bench ♥ 438 · Miguerutb/allenai-OLMoE-1B-7B-0924 ♥ 1

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 4.2 GB at INT4, 8.3 GB at FP8, 17 GB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

3 dated articles and 30 repo commits so far

Teal is the weights (2024-07-20), amber triangles are 3 dated articles (each links to it), grey ticks are repo commits.

What people built on it

20 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

20 derivatives on the Hub, led by 9 quantized

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF allenai/OLMoE-1B-7B-0924-GGUF for llama.cpp, Ollama, LM Studio 14.4k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

6 rows

Benchmarks 6 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (6)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.

Benchmark	Value	Setting / notes	Reported by	Date
author-reported ARC Challenge · Chat and follow instructions	62.1 percent	OLMoE-1B-7B	model author	2026-10-07
author-reported Commonsense QA sets · Chat and follow instructions	84.2 percent	OLMoE-1B-7B	model author	2026-10-07
author-reported Commonsense QA sets · Chat and follow instructions	79.8 percent	OLMoE-1B-7B	model author	2026-10-07
author-reported HellaSwag · Chat and follow instructions	80 percent	OLMoE-1B-7B	model author	2026-10-07
author-reported MMLU · Reason through hard problems	54.1 percent	OLMoE-1B-7B	model author	2026-10-07
author-reported Winogrande · Chat and follow instructions	70.2 percent	OLMoE-1B-7B	model author	2026-10-07

What the card says

1 cited sentence

Summary generated, cited

OLMoE-1B-7B is a Mixture-of-Experts LLM with 6.92B total parameters and 1B active parameters.^[1]^[2]

“OLMoE-1B-7B is a Mixture-of-Experts LLM with 1B active an…”
parameters human

Generated 2026-10-07 15:13 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

3 use cases

What it is good at 3 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Chat and follow instructionsauthor-reportedAuthor reports HellaSwag 80.0 Author reports ARC-Chall. 62.1 Author reports ARC-Easy 84.2 +1 more
Reason through hard problemsauthor-reportedAuthor reports MMLU 54.1
Run on a laptop or phonestructuralname: 1B fits: single-gpu

Facts

23 fields

Listed in LLMs · MoE

Facts

OLMoE-1B-7B-0924 has 6.92B parameters. (safetensors metadata, observed 2026-10-24)
OLMoE-1B-7B-0924 is a mixture-of-experts model with 64 experts, routing 8 per token. (config.json, observed 2026-10-24)
Context length is 4,096 tokens (4K). (config.json max_position_embeddings, observed 2026-10-24)
License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
Created 2024-07-20, last modified 2024-10-19. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	OlmoeForCausalLM
Model type	olmoe
Dtype	bfloat16
Params by dtype	BF16: 6,919,161,856
Size band	M
MoE	yes
Experts	64
Experts per token	8
Context length	4,096 (config.json max_position_embeddings)
Languages	en
Papers	arXiv:2409.02060
License	Apache 2.0 (apache-2.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · OlmoeForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · OlmoeForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for OLMoE-1B-7B-0924: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	17 GB · Single GPU (<24 GB)
FP8 (1 byte/param)	8.3 GB · Single GPU (<24 GB)
INT4 (0.5 bytes/param)	4.2 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.

All variants

20 variants

Lineage

Original model Family: allenai/OLMoE 20 derivatives on the Hub

finetune: 7quantized: 9adapter: 4

Top variants (20 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
allenai/MolmoE-1B-0924	finetune	—	—	—	apache-2.0	allenai	159	247.9k
allenai/OLMoE-1B-7B-0924-SFT	finetune	—	6.92B	M	apache-2.0	allenai	19	27.1k
allenai/OLMoE-1B-7B-0924-GGUF	quantized	GGUF	—	—	apache-2.0	allenai	10	14.4k
mradermacher/OLMoE-1B-7B-0924-i1-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	0	11k
mradermacher/OLMoE-1B-7B-0924-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	0	4.2k
tensorblock/OLMoE-1B-7B-0924-GGUF	quantized	GGUF	—	—	apache-2.0	tensorblock	0	2.7k
allura-org/MoE-Girl-1BA-7BT	finetune	—	6.92B	M	apache-2.0	allura-org	21	1k
DevQuasar-4/OLMoE-1B-7B-0924-GGUF	quantized	GGUF	—	—	—	DevQuasar-4	0	1k
sqlinn/uq-OLMoE-1B-7B-0924	finetune	—	6.92B	M	apache-2.0	sqlinn	1	597
Elmermoreno/OLMoE-1B-7B-0924-Q4_K_M-GGUF	quantized	GGUF	—	—	apache-2.0	Elmermoreno	0	200
Elmermoreno/OLMoE-1B-7B-0924-Q4_0-GGUF	quantized	Q4_0	—	—	apache-2.0	Elmermoreno	0	173
Kieran2828/OLMoE-1B-7B-0924-Q4_0-GGUF	quantized	Q4_0	—	—	apache-2.0	Kieran2828	0	149
leeminwaan/olmoe-reasoning-v1	adapter	—	—	—	—	leeminwaan	0	117
philipp-zettl/MolmoE-1B-0924	finetune	—	—	—	apache-2.0	philipp-zettl	0	88
anpaurehf/stego-olmoe-router-code	adapter	—	—	—	other	anpaurehf	0	74
AManzoni/olmoe_gsm8k_s42_hotk16	adapter	—	—	—	—	AManzoni	0	70
Xananthium/OLMoE-1B-7B-0924-INT8-W8A8	quantized	INT8	6.92B	M	apache-2.0	Xananthium	0	63
ffurfaro/Titans-v2-OLMoE-1B-7B-0924	finetune	—	—	—	apache-2.0	ffurfaro	1	0
leeminwaan/olmoe-reasoning-v2	adapter	—	—	—	—	leeminwaan	0	43
JIRU-W/top1moe-0.6b	finetune	—	6.92B	M	apache-2.0	JIRU-W	0	37

Repository changelog

30 commits

Repository changelog 30 commits

Date	Commit	SHA
2024-10-19	Update README.md	6d84c48
2024-09-14	Update README.md	4fa3a6e
2024-09-04	Update README.md	989ab78
2024-09-04	Update README.md	743d75a
2024-09-03	Update README.md	810676a
2024-09-03	Update README.md	cd7d397
2024-09-03	Update README.md	e4c66e0
2024-09-03	Update README.md	ff42442
2024-09-03	Update README.md	e8b61cb
2024-09-03	Update README.md	a10eca5
2024-09-03	Update README.md	bd8f786
2024-09-03	Update README.md	95e9d10
2024-09-03	Update README.md	3a97019
2024-09-03	Update README.md	be2a7c5
2024-09-03	Update README.md	0914349
2024-09-01	08 -> 09	bd1c52f
2024-08-30	Fix load balancing coefficient	83792c6
2024-08-28	Update README.md	4e1dee5
2024-08-28	Upload olmoe-logo.png	4c5204e
2024-08-16	Update README.md	fba02cf
2024-08-13	Update README.md	423c0f8
2024-08-13	Update README.md	ed327d8
2024-08-13	Update README.md	8f8468c
2024-08-13	Clarify branches	e775da2
2024-08-13	Move anneal branch to main	f8770d2
2024-08-09	Update README.md	1ae71b7
2024-08-09	Create README.md	66db7e7
2024-07-25	Delete model-00006-of-00006.safetensors	004807c
2024-07-25	Delete model-00005-of-00006.safetensors	b433fdd
2024-07-24	Update generation_config.json	a42569a

Articles and coverage

3 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official1

Ai2 OLMoE: an on-device open-source model

allenai.org2025-02-12officialauthorityabout the OLMoE familyHN 5 pts · 1 comments643 words

Ai2 released a fully open iOS app for the OLMoE model, allowing users to run AI privately on their devices. The app requires an iPhone 15 Pro or newer, or an M-series iPad, to function due to hardware constraints. The optimized stack achieves 41 tokens/s on average on an iPhone 16 Pro using Q4_K_M quantization. The model version allenai/OLMoE-1B-7B-0125-Instruct is 35% better on the evaluation suite while maintaining efficiency.written, reviewed, cited

The model uses the Dolmino mix for mid-training and the Tülu 3 post-training recipe.
Quantization causes IFEval scores to drop from 66.4 to 63.6 with minimal performance impact.
The app is built on top of Swift bindings for Llama.cpp to achieve on-device speed.

Analysis2

OLMoE and the hidden simplicity in training better foundation models

interconnects.ai2024-09-04authorityabout the OLMoE familyHN 3 pts · 0 comments2,058 words

Ai2 releases OLMoE, an open-source Mixture-of-Experts model with 1.3 billion active parameters and 6.9 billion total parameters. The model was trained on 5 trillion tokens and distributed under the Apache 2.0 license for all variants. It outperforms all open models in its active parameter range while remaining competitive with larger MoE architectures.written, reviewed, cited

Top labs allocate roughly 60% of compute to pretraining researchers, 25% to post-training, 10% to data, and 5% to other tasks.

An open, on-device model powered chat appfor iOS

apps.apple.com2025-02-11about the OLMoE familyHN 2 pts · 0 comments

Papers

1 paper

Papers 1

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

OLMoE: Open Mixture-of-Experts Language Models this model

Niklas Muennighoff, Luca Soldaini, Dirk Groeneveld +21 · 2024-09-03 · PDF · HF paper page (82 upvotes)

Similar models

3 models

Similar models

granite-4.0-h-tiny · 6.94B
same size band (M), also MoE, same task (text-generation), permissive license
LFM2.5-8B-A1B · 8.47B
same size band (M), also MoE, same task (text-generation)
Phi-tiny-MoE-instruct · 3.76B
same size band (M), also MoE, same task (text-generation), permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:41 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:13 UTC · Last verified 2026-10-25

Follow OLMoE-1B-7B-0924

Watch this model

/feeds/models/allenai--OLMoE-1B-7B-0924.xml(RSS 2.0) · every change to OLMoE-1B-7B-0924
/feeds/families/allenai--OLMoE.xml(RSS 2.0) · every change across the OLMoE family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.