Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Models › Cosmos family

Cosmos-Reason1-7B

Hugging Face ↗ Follow changes

nvidia NVIDIA Open Model License (nvidia-open-model-license) out 18 months ago

8.29B parameters, images and text in, text out, 125K tokens of context.

Verified 2026-10-25 Rank #735 in LLMs · score 71.6 method v1 Sources 3 Downloads 30d 17.3k · #468 of 619 tracked

Can I run it?5.0GB at INT4Estimatefits a 16 GB cardmethod
Can I use it?NVIDIA Open Model License (nvidia-open-model-license)Attribution requiredlicence file2026-10-24
Do people use it?17.3kdownloads, 30 days#468 of 619 we track · below the median+0 likes, downloads −4% this week248 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.

Most discussed · 259 HN pts

A Model of the Cosmos in the Ancient Greek Antikythera Mechanism

A new model for the Antikythera Mechanism explains the front display using a ring system with nine outputs.

  • The mechanism incorporates Babylonian period relations that are too inaccurate for naked-eye astronomy but fit the gear constraints.
  • Ancient designers derived planetary periods using a method from Plato's Parmenides dialogue rather than standard iterative approximation.
nature.com 2021-03-12 259 HN pts · 130 comments 6.7k words about the Cosmos family family
Latest · 5 months ago

Nvidia Cosmos 3

NVIDIA Cosmos 3 is an open foundation model for physical AI that unifies reasoning, world generation, and action generation within a single Mixture-of-Transformers (MoT) architecture.

  • The generator tower uses a diffusion-based process to generate physics-aware video and action outputs.
  • The Cosmos 3 Reasoner NIM is available today, while the full generation capabilities require the Generator NIM.
developer.nvidia.com 2026-06-01 150 HN pts · 28 comments 2.1k words about the Cosmos family family

3 best of 3 tracked

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

  1. ArticleUsing fluids to model inaccessible realms of the cosmosquantamagazine.org · 2023-07-15 · 97 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: 7 quantized builds and 4 Spaces.

  • As a service0no API provider listed
  • On own hardware7quantized builds
  • Built on77 fine-tunes · 0 adapters · 0 merges
  • In apps4Hugging Face Spaces

Top Spaces running it: prithivMLmods/DocScope-R1 ♥ 168 · hugging-apps/cmd-i2v-demo ♥ 2

Will it fit?

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

Fits a 16 GB card at INT4

About 5.0 GB at INT4, 10 GB at FP8, 20 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 5.0 GB of weights (estimate; KV cache and activations extra)INT45.0 GBFP8: about 10 GB of weights (estimate; KV cache and activations extra)FP810 GBBF16: about 20 GB of weights (estimate; KV cache and activations extra)BF1620 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 5.0 GB of weights (estimate; KV cache and activations extra)INT45.0 GBFP8: about 10 GB of weights (estimate; KV cache and activations extra)FP810 GBBF16: about 20 GB of weights (estimate; KV cache and activations extra)BF1620 GB

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

The story so far

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

Most coverage came in the first two weeks

Teal is the weights (2025-04-18), amber triangles are 3 dated articles (each links to it), grey ticks are repo commits.

AprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2025-12-10: 1 repo commit2025-08-14: 1 repo commit2025-08-06: 2 repo commits2025-08-01: 2 repo commits2025-06-11: 2 repo commits2025-05-27: 2 repo commits2025-05-23: 1 repo commit2025-05-20: 1 repo commit2025-05-19: 1 repo commit2025-05-17: 1 repo commit2025-05-16: 2 repo commits2025-05-15: 1 repo commit2025-05-02: 2 repo commitsNvidia Cosmos 3 · 2026-06-01A Model of the Cosmos in the Ancient Greek Antikythera Mechanism · 2021-03-12Using fluids to model inaccessible realms of the cosmos · 2023-07-15Weights 2025-04-18 · 2025-04-18Weights 2025-04-18now
AprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2025-12-10: 1 repo commit2025-08-14: 1 repo commit2025-08-06: 2 repo commits2025-08-01: 2 repo commits2025-06-11: 2 repo commits2025-05-27: 2 repo commits2025-05-23: 1 repo commit2025-05-20: 1 repo commit2025-05-19: 1 repo commit2025-05-17: 1 repo commit2025-05-16: 2 repo commits2025-05-15: 1 repo commit2025-05-02: 2 repo commitsNvidia Cosmos 3 · 2026-06-01A Model of the Cosmos in the Ancient Greek Antikythera Mechanism · 2021-03-12Using fluids to model inaccessible realms of the cosmos · 2023-07-15Weights 2025-04-18 · 2025-04-18Weights 2025-04-18now

What people built on it

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

14 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.

7 quantized (Hugging Face model tree)7 fine-tunes (Hugging Face model tree)7 quantized7 fine-tunes
7 quantized (Hugging Face model tree)7 fine-tunes (Hugging Face model tree)7 quantized7 fine-tunes
GGUF unsloth/Cosmos-Reason1-7B-GGUF for llama.cpp, Ollama, LM Studio 58.3k downloads MLX introvoyz041/Cosmos-Reason1-7B-mlx-4Bit for Apple silicon 142 downloads

All variants →

Reference

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

Benchmarks

7 rows

Benchmarks 7 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (7)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.
BenchmarkValueSetting / notesReported byDate
author-reported AV70.8 percent—model author2026-10-07
author-reported Agibot48.9 percent—model author2026-10-07
author-reported Average65.1 percentAccuracymodel author2026-10-07
author-reported BridgeDataV263.7 percent—model author2026-10-07
author-reported HoloAssist62.7 percent—model author2026-10-07
author-reported RoboFail57.2 percent—model author2026-10-07
author-reported RoboVQA87.3 percent—model author2026-10-07

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.

Facts

22 fields

Listed in LLMs

Facts

Pipeline tagimage-text-to-text
Librarytransformers
ArchitecturesQwen2_5_VLForConditionalGeneration
Model typeqwen2_5_vl
Dtypebfloat16
Params by dtypeBF16: 8,292,166,656
Size bandM
MoEno
Context length128,000 (config.json max_position_embeddings)
Languagesen
PapersarXiv:2503.15558
LicenseNVIDIA Open Model License (nvidia-open-model-license) (other)
Hub snapshot2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Qwen2_5_VLForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen2_5_VLForConditionalGeneration · 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 Cosmos-Reason1-7B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)20 GB · Single GPU (<24 GB)
FP8 (1 byte/param)10 GB · Single GPU (<24 GB)
INT4 (0.5 bytes/param)5.0 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

14 variants

Lineage

Fine-tuned from Qwen/Qwen2.5-VL-7B-Instruct Family: nvidia/Cosmos 14 derivatives on the Hub

Base model: Qwen/Qwen2.5-VL-7B-Instruct

quantized: 7finetune: 7

Top variants (14 shown, ranked by downloads)

Relation
Quant
14 rows
Variant Relation Quant Params Band License Author Likes Downloads
unsloth/Cosmos-Reason1-7B-GGUF quantized GGUF — — other unsloth 14 58.3k
mradermacher/Cosmos-Reason1-7B-i1-GGUF quantized GGUF — — other mradermacher 1 15.6k
mradermacher/Cosmos-Reason1-7B-GGUF quantized GGUF — — other mradermacher 1 8.6k
DevQuasar/nvidia.Cosmos-Reason1-7B-GGUF quantized GGUF — — — DevQuasar 3 3.3k
unsloth/Cosmos-Reason1-7B finetune — 8.29B M other unsloth 2 1.4k
unsloth/Cosmos-Reason1-7B-unsloth-bnb-4bit quantized -bnb 8.50B M other unsloth 0 1.5k
unsloth/Cosmos-Reason1-7B-bnb-4bit quantized -bnb 8.52B M other unsloth 1 893
Zaixi/STELLA-VLM-FineBio-7B finetune — 8.29B M mit Zaixi 4 69
sasa2000/Cosmos-Reason1-7B-heretic finetune — 8.29B M other sasa2000 2 138
Zaixi/STELLA-VLM-JoVE-7B finetune — 8.29B M mit Zaixi 3 17
introvoyz041/Cosmos-Reason1-7B-mlx-4Bit quantized mlx 7.62B M other introvoyz041 0 142
noebreton/qwen2-7b-instruct-trl-sft-ChartQA finetune — — — — noebreton 0 0
noebreton/cosmos_custom_legacy_prompt finetune — — — — noebreton 0 0
noebreton/cosmos_custom_one_prompt finetune — — — — noebreton 0 0

Repository changelog

19 commits

Repository changelog 19 commits

DateCommitSHA
2025-12-10 Update README.md 375e240
2025-08-14 Improve metadata 🤗 (#3) 3210bec
2025-08-06 Update README.md 4b2bdaa
2025-08-06 Update README.md 25940bb
2025-08-01 Update README.md 8fe96c1
2025-08-01 08/01 release: Added support for spatial-temporal reasoning of city and industrial operations 0caf724
2025-06-11 Update README.md 1674a72
2025-06-11 Updated checkpoint with better critic capability 2464fff
2025-05-27 Update README.md 5c3a31c
2025-05-27 Update README.md 5cace25
2025-05-23 Update README.md 6554ea3
2025-05-20 Fix library name (#1) b0b77be
2025-05-19 Update README.md 19561cf
2025-05-17 Update README.md c786949
2025-05-16 Add generation_config.json df59fbe
2025-05-16 Update README.md 260e2ec
2025-05-15 Latest checkpoint for release 098a5bb
2025-05-02 Create README.md 01f631b
2025-05-02 Latest RL checkpoint 4e92fc4

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

Nvidia Cosmos 3
developer.nvidia.com2026-06-01officialauthorityHN front pageabout the Cosmos familyHN 150 pts · 28 comments2,118 words
NVIDIA Cosmos 3 is an open foundation model for physical AI that unifies reasoning, world generation, and action generation within a single Mixture-of-Transformers (MoT) architecture. The model supports modalities including Text to Image, Text to Video, and Action to Video outputs. It offers a 16B parameter Nano model for workstation inference and a 64B parameter Super model for datacenter deployment.written, reviewed, cited
  • The generator tower uses a diffusion-based process to generate physics-aware video and action outputs.
  • The Cosmos 3 Reasoner NIM is available today, while the full generation capabilities require the Generator NIM.

Analysis2

A Model of the Cosmos in the Ancient Greek Antikythera Mechanism
nature.com2021-03-12HN front pageabout the Cosmos familyHN 259 pts · 130 comments6,686 words
A new model for the Antikythera Mechanism explains the front display using a ring system with nine outputs. Inscriptions reveal factorizable period relations for Venus (289, 462) and Saturn (427, 442), which allow mechanization with gears under 100 teeth, unlike the prime-numbered Babylonian periods.written, reviewed, cited
  • The mechanism incorporates Babylonian period relations that are too inaccurate for naked-eye astronomy but fit the gear constraints.
  • Ancient designers derived planetary periods using a method from Plato's Parmenides dialogue rather than standard iterative approximation.
  • The front display uses nested tubes with arms supporting rings to carry nine specific astronomical outputs.
Using fluids to model inaccessible realms of the cosmos
quantamagazine.org2023-07-15about the Cosmos familyHN 97 pts · 78 comments2,219 words
Silke Weinfurtner leads analogue gravity research at the University of Nottingham, utilizing fluid systems to simulate inaccessible cosmic phenomena like black holes and the early universe. Since the 1970s, such experiments have mapped sound waves in fluids to Hawking radiation and used preheating mechanisms to model cosmic inflation seeds. These setups rely on Navier-Stokes equations that, under specific assumptions, reduce to forms describing quantum fields in curved space-time.written, reviewed, cited
  • The experiment uses a vortex in an oversize bathtub to create an analogue of a black hole's inescapable event horizon.
  • Researchers simulate cosmic inflation by driving surface wave speed toward zero so the wave never reaches the boundary.
  • Analogue experiments test cosmological approximations by embedding them in a physical reality with all interaction terms built in.

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).

Similar models

3 models

Similar models

  • Qwen3-VL-8B-Instruct · 8.77B
    same size band (M), same task (image-text-to-text), more providers, permissive license
  • Qwen2.5-VL-7B-Instruct · 8.29B
    same size band (M), same task (image-text-to-text), more providers, permissive license
  • Qwen3.5-9B · 9.65B
    same task (image-text-to-text), more providers, 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 Cosmos-Reason1-7B

Watch this model

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