The story so far
Weights, every dated article, and repo commits on one rail.
6 repo commits since the weights landed
Teal is the weights (2026-07-13), grey ticks are repo commits.
OvisOCR2
ATH-MaaS Apache 2.0 out 3 months ago
853M parameters, images and text in, text out, 256K tokens of context. 1 provider serves it.
Verified 2026-10-25 Rank #452 in LLMs · score 78 method v1 Sources 3 Downloads 30d 135.7k · #284 of 619 tracked
Changed this week: 2 new quantized builds on the Hub (26 in all).2026-10-24
Do people use it?135.7kdownloads, 30 days#284 of 619 we track · top 46%+8 likes, downloads −51% this week502 likesHugging Face2026-10-24
Where to run it
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
| # | Cheapest live row | In /1M | Out /1M | Context | Quant | Seen |
|---|---|---|---|---|---|---|
| 1 | Featherless | — | — | — | — | Oct 25 |
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.
Served by 1 API provider; 26 quantized builds for local use.
- As a service1Featherless
- On own hardware26quantized builds
- Built on33 fine-tunes · 0 adapters · 0 merges
- In apps12Hugging Face Spaces
Top Spaces running it: ATH-MaaS/OvisOCR2 ♥ 117 · hugging-apps/ovis-ocr2-demo ♥ 6 · Tonic/ocr-ebola ♥ 1
Will it fit?
Weights only, an estimate: leave room for the KV cache. Method
Fits a 16 GB card at INT4
About 0.5 GB at INT4, 1.0 GB at FP8, 2.0 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
29 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
29 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.
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 overall (PaddlePaddle/Real5-OmniDocBench) | 92.29 | — | Real5-OmniDocBench Leaderboard | 2026-10-24 |
| author-reported scanning (PaddlePaddle/Real5-OmniDocBench) | 93.77 | — | Real5-OmniDocBench Leaderboard | 2026-10-24 |
| author-reported warping (PaddlePaddle/Real5-OmniDocBench) | 91.4 | — | Real5-OmniDocBench Leaderboard | 2026-10-24 |
| author-reported screen_photography (PaddlePaddle/Real5-OmniDocBench) | 93.09 | — | Real5-OmniDocBench Leaderboard | 2026-10-24 |
| author-reported illumination (PaddlePaddle/Real5-OmniDocBench) | 92.88 | — | Real5-OmniDocBench Leaderboard | 2026-10-24 |
| author-reported skew (PaddlePaddle/Real5-OmniDocBench) | 90.33 | — | Real5-OmniDocBench Leaderboard | 2026-10-24 |
What the card says
1 cited sentence
Summary generated, cited
The model generates Markdown representations of document pages, including text, formulas, and tables.[1]
Generated 2026-10-07 15:09 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
Use-case evidence
7 use cases
What it is good at 7 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=0.96tag: multimodalpipeline: image-text-to-text+1 more
- Extract text from documentsstructuraldecision-model estimate p=1.00name: OCRtag: ocr
- Chat and follow instructionsstructuraltag: conversationalchat template present
- Run on a laptop or phonestructuraldecision-model estimate p=0.99fits: edge
- Call tools and APIsstructuralchat template: <tool_call>
- Reason through hard problemsstructuralchat template: <think>
- Read long documentsstructuralconfig: 262,144 context
Facts
21 fields
Listed in LLMs
Facts
- OvisOCR2 has 853M parameters. (safetensors metadata, observed 2026-10-24)
- Context length is 262,144 tokens (256K). (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 2026-07-13, last modified 2026-10-22. (HF API, observed 2026-10-24)
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | Qwen3_5ForConditionalGeneration |
| Model type | qwen3_5 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 852,985,920 |
| Size band | S |
| MoE | no |
| Context length | 262,144 (config.json max_position_embeddings) |
| Papers | arXiv:2607.13639 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
Servability
vLLM · SGLang
Servability
vLLM: supported · Qwen3_5ForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3_5ForConditionalGeneration · 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 OvisOCR2: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 2.0 GB · Edge (<4 GB) |
|---|---|
| FP8 (1 byte/param) | 1.0 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 0.5 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
All variants
29 variants
Lineage
Base model: Qwen/Qwen3.5-0.8B
quantized: 26finetune: 3
Top variants (29 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| Abiray/OvisOCR2-GGUF | quantized | GGUF | — | — | apache-2.0 | Abiray | 28 | 30.3k |
| bartowski/ATH-MaaS_OvisOCR2-GGUF | quantized | GGUF | — | — | apache-2.0 | bartowski | 4 | 14.8k |
| prithivMLmods/OvisOCR2-F32-GGUF | quantized | GGUF | — | — | apache-2.0 | prithivMLmods | 4 | 5.7k |
| MDfox/OvisOCR2-GGUF-SOTA-20260823 | quantized | GGUF | — | — | apache-2.0 | MDfox | 1 | 4k |
| mlx-community/OvisOCR2-8bit | quantized | -8bit | 853M | S | apache-2.0 | mlx-community | 2 | 1.6k |
| Disya/OvisOCR2-GGUF | quantized | GGUF | — | — | — | Disya | 1 | 1.5k |
| manojbillionaire123/ATH-MaaS_OvisOCR2-GGUF | quantized | GGUF | — | — | apache-2.0 | manojbillionaire123 | 0 | 1.1k |
| sahilchachra/ovisocr2-mxfp8-mlx | quantized | fp8- | 853M | S | apache-2.0 | sahilchachra | 2 | 726 |
| enacimie/OvisOCR2-Q8_0-GGUF | quantized | Q8_0 | — | — | apache-2.0 | enacimie | 0 | 406 |
| Ar4ikov/OvisOCR2-AWQ | quantized | AWQ | 1.12B | S | apache-2.0 | Ar4ikov | 2 | 290 |
| mlx-community/OvisOCR2-6bit | quantized | — | 853M | S | apache-2.0 | mlx-community | 1 | 333 |
| enacimie/OvisOCR2-Q4_K_M-GGUF | quantized | GGUF | — | — | apache-2.0 | enacimie | 0 | 372 |
| sahilchachra/ovisocr2-int8-mlx | quantized | int8 | 853M | S | apache-2.0 | sahilchachra | 1 | 312 |
| st0722/OvisOCR2-MLX-BF16 | finetune | MLX | 853M | S | apache-2.0 | st0722 | 0 | 362 |
| sahilchachra/ovisocr2-int4-mlx | quantized | int4 | 853M | S | apache-2.0 | sahilchachra | 1 | 284 |
| mlboydaisuke/OvisOCR2-LiteRT | quantized | — | — | — | apache-2.0 | mlboydaisuke | 1 | 171 |
| AlioLeuchtmann/receipt-to-json-models | finetune | — | 752M | S | apache-2.0 | AlioLeuchtmann | 1 | 159 |
| st0722/OvisOCR2-MLX-4bit | quantized | MLX | 853M | S | apache-2.0 | st0722 | 0 | 200 |
| mlx-community/OvisOCR2-4bit | quantized | -4bit | 853M | S | apache-2.0 | mlx-community | 0 | 198 |
| yugeshkarunamurthy/OvisOCR2-oQ4e | quantized | — | 853M | S | apache-2.0 | yugeshkarunamurthy | 1 | 113 |
| sahilchachra/ovisocr2-mxfp4-mlx | quantized | mxfp4 | 853M | S | apache-2.0 | sahilchachra | 0 | 141 |
| tokimoa/ovisocr2-mlx-bf16 | finetune | mlx | 853M | S | apache-2.0 | tokimoa | 0 | 114 |
| yugeshkarunamurthy/OvisOCR2-oQ6e | quantized | — | 853M | S | apache-2.0 | yugeshkarunamurthy | 0 | 90 |
| tokimoa/ovisocr2-mlx-8bit | quantized | mlx | 853M | S | apache-2.0 | tokimoa | 0 | 90 |
| tokimoa/ovisocr2-mlx-4bit | quantized | mlx | 853M | S | apache-2.0 | tokimoa | 0 | 85 |
| onnx-community/OvisOCR2-ONNX | quantized | — | — | — | — | onnx-community | 1 | 0 |
| parkingsoman/ovisocr2-onnx-webgpu | quantized | — | — | — | apache-2.0 | parkingsoman | 0 | 28 |
| vlm-run/OvisOCR2-NVFP4 | quantized | NVFP4 | 604M | S | apache-2.0 | vlm-run | 0 | 2 |
| cstr/OvisOCR2-GGUF | quantized | GGUF | — | — | apache-2.0 | cstr | 0 | 0 |
Repository changelog
6 commits
changed in the last 7 daysRepo updated: 2026-10-23Repo commits: 2026-10-22Other: 2026-10-24all changes
Repository changelog 6 commits
| Date | Commit | SHA |
|---|---|---|
| 2026-10-22 | Update README.md | cbdede8 |
| 2026-08-28 | Add Real5-OmniDocBench evaluation results for OvisOCR2 (#11) | 1fc9221 |
| 2026-07-16 | Update README.md | 65c619d |
| 2026-07-15 | add technical report and online demo links | d844a45 |
| 2026-07-13 | Upload folder using huggingface_hub | 77bfe94 |
| 2026-07-13 | initial commit | 666854c |
All providers
1 provider · 1 endpoint
Where to run 1 providers · 1 endpoints
Hosted endpoints observed for OvisOCR2, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Featherless hf | ATH-MaaS/OvisOCR2 | — | — | — | — | — | — | — | live | 2026-10-25 |
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).
- OvisOCR2 Technical Report this model
Similar models
3 models
Similar models
- Qwen3.5-0.8B · 873M same size band (S), same task (image-text-to-text), permissive license
- Florence-2-large · 777M same size band (S), same task (image-text-to-text), permissive license
- Qwen3.5-0.8B-Base · 873M same size band (S), same task (image-text-to-text), permissive license
Sources
3 sources
Sources
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:39 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:09 UTC · Last verified 2026-10-25
Follow OvisOCR2
Watch this model
- /feeds/models/ATH-MaaS--OvisOCR2.xml(RSS 2.0) · every change to OvisOCR2
- /feeds/families/ATH-MaaS--OvisOCR2.xml(RSS 2.0) · every change across the OvisOCR2 family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.