Solar-Open-100B

upstage upstage-solar-license out 11 months ago

103B parameters (8 of 128 experts per token), text in, text out, 128K tokens of context.

Verified 2026-10-25 Rank #661 in LLMs · score 73 method v1 Sources 3 Downloads 30d 21.5k · #450 of 619 tracked

Can I run it?62GB at INT4Estimatefits a 64 GB Mac (unified memory)method

Can I use it?upstage-solar-licenseCustom license; read the repo LICENSE before commercial hostinglicence file2026-10-24

Do people use it?21.5kdownloads, 30 days#450 of 619 we track · below the median497 likesHugging Face2026-10-24

Worth reading

Picked by rule from 2 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Most discussed · 936 HN pts

If the moon were only 1 pixel: A tediously accurate solar system model (2014)

Traveling to Jupiter at 75mi/hr would take over 500 years from Earth.
The New Horizons spacecraft launched in 2006 and reached Jupiter in 13 months.

joshworth.com 2025-06-13 936 HN pts · 261 comments 1.3k words about the Solar family family

2 best of 2 tracked

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

DiscussionShow HN: J.E.S.S.– Open-Source Orbital AI Supercluster(Solar Swarms,1.3 GW Ring)news.ycombinator.com · 2025-11-15 · 1 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: 14 quantized builds and 7 Spaces.

As a service0no API provider listed
On own hardware14quantized builds
Built on66 fine-tunes · 0 adapters · 0 merges
In apps7Hugging Face Spaces

Top Spaces running it: mayafree/Model-Genome-Korea ♥ 33

Will it fit?

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

Fits a 64 GB Mac (unified memory) at INT4

About 62 GB at INT4, 123 GB at FP8, 246 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.

2 dated articles and 24 repo commits so far

Teal is the weights (2025-12-10), amber triangles are 2 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, most of them quantized

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

GGUF mradermacher/Solar-Open-100B-i1-GGUF for llama.cpp, Ollama, LM Studio 21.1k downloads MLX mlx-community/Solar-Open-100B-4bit for Apple silicon 828 downloads NVFP4 Firworks/Solar-Open-100B-nvfp4 for vLLM, SGLang 29.9k downloads

All variants →

Reference

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

Benchmarks

1 row

Benchmarks 1 row

measured third-party leaderboard rowauthor-reported number from the model card or eval files (1)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 MMLU · Reason through hard problems	73	—	model author	2026-10-07

Use-case evidence

4 use cases

What it is good at 4 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.

Reason through hard problemsauthor-reportedAuthor reports KMMLU 73.0 chat template: <|think|>
Chat and follow instructionsstructuraldecision-model estimate p=0.84 tag: conversational chat template present
Read long documentsstructuraldecision-model estimate p=1.00 config: 131,072 context
Call tools and APIsstructuralchat template: tool_calls

Facts

24 fields

Listed in LLMs · MoE

Facts

Solar-Open-100B has 103B parameters. (safetensors metadata, observed 2026-10-24)
Solar-Open-100B is a mixture-of-experts model with 128 experts, routing 8 per token. (config.json, observed 2026-10-24)
Context length is 131,072 tokens (128K). (config.json max_position_embeddings, observed 2026-10-24)
License: upstage-solar-license (friction: review). Custom license; read the repo LICENSE before commercial hosting. (license file, observed 2026-10-24)
Created 2025-12-10, last modified 2026-01-30. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	SolarOpenForCausalLM
Model type	solar_open
Dtype	bfloat16
Params by dtype	BF16: 102,651,793,408 · F32: 6,144
Size band	XXL
MoE	yes
Experts	128
Experts per token	8
Context length	131,072 (config.json max_position_embeddings)
Languages	en, ko
Papers	arXiv:2601.07022
License	upstage-solar-license (other)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: not in registry · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 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 Solar-Open-100B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	246 GB · Multi-GPU node (<640 GB)
FP8 (1 byte/param)	123 GB · Multi-GPU node (<640 GB)
INT4 (0.5 bytes/param)	62 GB · Single high-end GPU (<80 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: upstage/Solar 20 derivatives on the Hub

quantized: 14finetune: 6

Top variants (20 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
Firworks/Solar-Open-100B-nvfp4	quantized	nvfp4	58.5B	XL	—	Firworks	5	29.9k
furiosa-ai/Solar-Open-100B-NVFP4A16	quantized	—	60.4B	XL	other	furiosa-ai	2	24.3k
mradermacher/Solar-Open-100B-i1-GGUF	quantized	GGUF	—	—	other	mradermacher	7	21.1k
AaryanK/Solar-Open-100B-GGUF	quantized	GGUF	—	—	other	AaryanK	16	17.2k
mradermacher/Solar-Open-100B-GGUF	quantized	GGUF	—	—	other	mradermacher	1	9.5k
DevQuasar-11/upstage.Solar-Open-100B-GGUF	quantized	GGUF	—	—	—	DevQuasar-11	0	5.9k
cyankiwi/Solar-Open-100B-AWQ-4bit	quantized	AWQ	18.1B	L	other	cyankiwi	1	2.6k
haanjack/Solar-Open-100B-MXFP4	quantized	MXFP4	57.4B	XL	apache-2.0	haanjack	0	1.6k
mlx-community/Solar-Open-100B-4bit	quantized	-4bit	103B	XXL	other	mlx-community	0	828
cyankiwi/Solar-Open-100B-AWQ-8bit	quantized	AWQ	30.6B	L	other	cyankiwi	2	525
mlx-community/Solar-Open-100B-8bit	quantized	-8bit	103B	XXL	other	mlx-community	0	565
litmudoc/Solar-Open-100B-MXFP4-Q8	quantized	MXFP4	103B	XXL	other	litmudoc	0	558
mlx-community/Solar-Open-100B-6bit	quantized	—	103B	XXL	other	mlx-community	0	447
hell0ks/Solar-Open-100B-jailbreak	finetune	—	103B	XXL	other	hell0ks	4	218
mike-ravkine/Solar-Open-100B-FP8-Dynamic	quantized	-FP8	103B	XXL	other	mike-ravkine	0	305
llaa33219/Solar-Open-100B-to-8B-test	finetune	—	7.77B	M	apache-2.0	llaa33219	0	237
llaa33219/Solar-Open-100B-to-4B-test-2	finetune	—	4.40B	M	apache-2.0	llaa33219	0	91
llaa33219/Solar-Open-100B-to-7B-test	finetune	—	7.21B	M	apache-2.0	llaa33219	0	49
llaa33219/Solar-Open-100B-pruned-5pct-test1	finetune	—	36.3B	XL	apache-2.0	llaa33219	0	44
llaa33219/Solar-Open-100B-pruned-5pct-test2	finetune	—	36.3B	XL	apache-2.0	llaa33219	0	37

Repository changelog

24 commits

Repository changelog 24 commits

Date	Commit	SHA
2026-01-30	Update transformers version to 5.0 (#24)	1f59143
2026-01-28	Add Nota AI quantization model, remove Public API Access section	a383611
2026-01-16	Update solar-open-technical-report.pdf from tmp	4ad167d
2026-01-16	Update LICENSE from tmp	d7331ce
2026-01-16	Update README.md from tmp	b87796b
2026-01-15	Link model card to technical report and project page (#22)	c90f8f3
2026-01-05	Update Pipeline_tag (#20)	e9e6866
2026-01-05	Adding `transformers` as the library name (#18)	0a57d6c
2026-01-05	Update README.md	82e8365
2026-01-05	add solar open technical report. add benchmark results. (#19)	ac16609
2026-01-05	Update config.json (#17)	55cb6b9
2026-01-02	Update README.md	c8988f1
2026-01-01	Update README.md	e4db2e9
2026-01-01	Fix the license of Solar modeling and inference code.	ff17f06
2025-12-31	Update README.md	c03acad
2025-12-31	Update README.md	fd4b56c
2025-12-31	Add files using upload-large-folder tool	b6e8096
2025-12-31	Add files using upload-large-folder tool	2754e05
2025-12-15	Update README: New intro, API date to Dec 31 2025, remove agentic claims	7a9b40d
2025-12-15	Update README: Remove agentic claims & update API docs	473f1f1
2025-12-13	Add clickable LICENSE links in README	113e7e0
2025-12-13	Add Solar-Apache License 2.0	c9fe5d7
2025-12-13	Add model card for Solar Open 100B	5751e6f
2025-12-10	initial commit	8efc246

Articles and coverage

2 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.

Analysis1

If the moon were only 1 pixel: A tediously accurate solar system model (2014)

joshworth.com2025-06-13HN front pageabout the Solar familyHN 936 pts · 261 comments1,336 words

A pixel-scale visualization illustrates that the solar system is overwhelmingly empty space, with distances between planets requiring 11 more maps to depict the average distance from a hydrogen proton to its electron. The known universe is 99.9999999999999999999958% empty, and viewing the entire scale would require scrolling through 6,771 more maps after the initial journey.written, reviewed, cited

Traveling to Jupiter at 75mi/hr would take over 500 years from Earth.
The New Horizons spacecraft launched in 2006 and reached Jupiter in 13 months.
Printing the map at 300 pixels per inch requires paper 475 feet wide.

Community1

Show HN: J.E.S.S.– Open-Source Orbital AI Supercluster(Solar Swarms,1.3 GW Ring)

news.ycombinator.com2025-11-15about the Solar familyHN 1 pts · 0 comments116 words

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).

Solar Open Technical Report this model

Sungrae Park, Sanghoon Kim, Jungho Cho +34 · 2026-01-11 · PDF · HF paper page (66 upvotes)

Similar models

3 models

Similar models

gpt-oss-120b · 117B
same size band (XXL), also MoE, same task (text-generation), more providers, permissive license
GLM-4.5-Air · 110B
same size band (XXL), also MoE, same task (text-generation), more providers, permissive license
Qwen3-Next-80B-A3B-Instruct · 81.3B
also MoE, same task (text-generation), more providers, permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:42 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:14 UTC · Last verified 2026-10-25

Follow Solar-Open-100B

Watch this model

/feeds/models/upstage--Solar-Open-100B.xml(RSS 2.0) · every change to Solar-Open-100B
/feeds/families/upstage--Solar.xml(RSS 2.0) · every change across the Solar family

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