The story so far
Weights, every dated article, and repo commits on one rail.
15 repo commits since the weights landed
Teal is the weights (2025-04-10), grey ticks are repo commits.
InternVL3-78B
OpenGVLab Qwen License (qwen) out 18 months ago
78.4B parameters, images and text in, text out.
Verified 2026-10-25 Rank #585 in LLMs · score 74.3 method v1 Sources 3 Downloads 30d 8.3k · #519 of 619 tracked
Do people use it?8.3kdownloads, 30 days#519 of 619 we track · below the median+0 likes, downloads +5% this week239 likesHugging Face2026-10-24
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: 4 quantized builds and 6 Spaces.
- As a service0no API provider listed
- On own hardware4quantized builds
- Built on22 fine-tunes · 0 adapters · 0 merges
- In apps6Hugging Face Spaces
Top Spaces running it: TIGER-Lab/MEGA-Bench ♥ 51
Will it fit?
Weights only, an estimate: leave room for the KV cache. Method
Fits a 64 GB Mac (unified memory) at INT4
About 47 GB at INT4, 94 GB at FP8, 188 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
What people built on it
6 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
6 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.
Reference
Everything else, folded: the full tables, every source.
What the card says
1 cited sentence
Use-case evidence
5 use cases
What it is good at 5 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.
- Understand images and videostructuraldecision-model estimate p=1.00tag: image-text-to-textpipeline: image-text-to-text+1 more
- Chat and follow instructionsstructuraldecision-model estimate p=0.99tag: conversationalchat template present
- Work across languagesstructuraldecision-model estimate p=0.71tag: multilingual
- Embed text for searchstructuraltag: feature-extraction
- Call tools and APIsstructuraldecision-model estimate p=0.88
Facts
19 fields
Listed in LLMs
Facts
- InternVL3-78B has 78.4B parameters. (safetensors metadata, observed 2026-10-24)
- License: Qwen License (qwen) (friction: gated-threshold). (license file, observed 2026-10-24)
- Created 2025-04-10, last modified 2025-09-11. (HF API, observed 2026-10-24)
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | InternVLChatModel |
| Model type | internvl_chat |
| Dtype | bfloat16 |
| Params by dtype | BF16: 78,408,318,336 |
| Size band | XL |
| MoE | no |
| Languages | multilingual |
| Papers | arXiv:2312.14238, arXiv:2404.16821, arXiv:2412.05271, arXiv:2411.10442, arXiv:2504.10479, arXiv:2412.09616 |
| License | Qwen License (qwen) (other) |
| Hub snapshot | 2026-10-25 |
Servability
vLLM · SGLang
Servability
vLLM: supported · InternVLChatModel · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · InternVLChatModel · 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 InternVL3-78B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 188 GB · Multi-GPU node (<640 GB) |
|---|---|
| FP8 (1 byte/param) | 94 GB · Multi-GPU node (<640 GB) |
| INT4 (0.5 bytes/param) | 47 GB · Single high-end GPU (<80 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
All variants
6 variants
Lineage
Base model: OpenGVLab/InternVL3-78B-Instruct
quantized: 4finetune: 2
Top variants (6 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| OpenGVLab/InternVL3-78B-AWQ | quantized | AWQ | — | — | other | OpenGVLab | 11 | 508.7k |
| mradermacher/InternVL3-78B-i1-GGUF | quantized | GGUF | — | — | other | mradermacher | 1 | 9.2k |
| unsloth/InternVL3-78B-GGUF | finetune | GGUF | — | — | other | unsloth | 3 | 8.7k |
| mradermacher/InternVL3-78B-GGUF | quantized | GGUF | — | — | other | mradermacher | 2 | 4.6k |
| brandonbeiler/InternVL3-78B-FP8-Dynamic | quantized | -FP8 | 78.4B | XL | mit | brandonbeiler | 0 | 3.2k |
| AI45Research/SafeWork-R1-InternVL3-78B | finetune | — | 78.4B | XL | apache-2.0 | AI45Research | 1 | 116 |
Repository changelog
15 commits
Repository changelog 15 commits
| Date | Commit | SHA |
|---|---|---|
| 2025-09-11 | Upload tokenizer_config.json with huggingface_hub | 4a42b16 |
| 2025-09-11 | Upload tokenizer_config.json with huggingface_hub | e30f625 |
| 2025-05-29 | Update README.md | 852965e |
| 2025-04-17 | Upload README.md with huggingface_hub | 4f7e3ac |
| 2025-04-17 | Update modeling_intern_vit.py (#2) | 8edce0d |
| 2025-04-16 | Upload modeling_intern_vit.py with huggingface_hub | 054916e |
| 2025-04-15 | Upload modeling_internvl_chat.py with huggingface_hub | c29058f |
| 2025-04-15 | Upload README.md with huggingface_hub | 656c36d |
| 2025-04-14 | Upload README.md with huggingface_hub | 9e3847e |
| 2025-04-12 | Upload README.md with huggingface_hub | 471c42a |
| 2025-04-12 | Upload README.md with huggingface_hub | a106564 |
| 2025-04-12 | Upload README.md with huggingface_hub | e6554f0 |
| 2025-04-12 | Upload README.md with huggingface_hub | 4223725 |
| 2025-04-10 | Upload folder using huggingface_hub | 62ea90c |
| 2025-04-10 | initial commit | d855900 |
Papers
6 papers
Papers 6
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).
- V2PE: Improving Multimodal Long-Context Capability of Vision-Language Models with Variable Visual Position Encoding cited in card peer-reviewed · measuredJunqi Ge, Ziyi Chen, Jintao Lin +4 · 2024-12-12 · IEEE/CVF International Conference on Computer Vision (ICCV) · PDF · HF paper page (1 upvotes) · evidence
Similar models
3 models
Similar models
- Qwen2.5-VL-72B-Instruct · 73.4B same size band (XL), same task (image-text-to-text), more providers
- NVLM-D-72B · 79.4B same size band (XL), same task (image-text-to-text)
- QVQ-72B-Preview · 73.4B same size band (XL), same task (image-text-to-text), more providers
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 InternVL3-78B
Watch this model
- /feeds/models/OpenGVLab--InternVL3-78B.xml(RSS 2.0) · every change to InternVL3-78B
- /feeds/families/OpenGVLab--InternVL3.xml(RSS 2.0) · every change across the InternVL3 family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.