Up to date
datalab-to OpenRAIL out 5 months ago
686M parameters, images and text in, text out, 256K tokens of context. 1 provider serves it.
Verified 2026-10-25 Rank #323 in LLMs · score 82.2 method v1 Sources 3 Downloads 30d 1.4M · #90 of 619 tracked
Picked by rule from 3 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Surya is a 650M parameter OCR model that performs layout analysis and table recognition in over 90 languages.
Surya is a 366M-parameter foundation model for heliophysics trained on full-resolution multi-instrument SDO observations across 13 channels from AIA and HMI instruments.
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.
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 |
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; 11 quantized builds for local use.
Top Spaces running it: build-small-hackathon/kirana-saathi ♥ 23 · build-small-hackathon/dukaan-saathi ♥ 21 · xiaoyao9184/surya ♥ 5 · davanstrien/the-recovered-page ♥ 5
Weights only, an estimate: leave room for the KV cache. Method
About 0.4 GB at INT4, 0.8 GB at FP8, 1.6 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2026-05-14), amber triangles are 3 dated articles (each links to it), grey ticks are repo commits.
19 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported internal benchmark set of 91 languages | 87.2% percent | — | model author | 2026-10-07 |
| author-reported olmOCR-bench · Extract text from documents | 83.3% percent | — | model author | 2026-10-07 |
| author-reported overall (allenai/olmOCR-bench) | 83.3 | — | Surya OCR 2 Model Card | 2026-10-24 |
| author-reported arxiv_math (allenai/olmOCR-bench) | 88.3 | — | Surya OCR 2 Model Card | 2026-10-24 |
| author-reported old_scans_math (allenai/olmOCR-bench) | 81.4 | — | Surya OCR 2 Model Card | 2026-10-24 |
| author-reported table_tests (allenai/olmOCR-bench) | 86.6 | — | Surya OCR 2 Model Card | 2026-10-24 |
| author-reported old_scans (allenai/olmOCR-bench) | 41.8 | — | Surya OCR 2 Model Card | 2026-10-24 |
| author-reported headers_footers (allenai/olmOCR-bench) | 92.5 | — | Surya OCR 2 Model Card | 2026-10-24 |
| author-reported multi_column (allenai/olmOCR-bench) | 82.4 | — | Surya OCR 2 Model Card | 2026-10-24 |
| author-reported long_tiny_text (allenai/olmOCR-bench) | 93.7 | — | Surya OCR 2 Model Card | 2026-10-24 |
| author-reported baseline (allenai/olmOCR-bench) | 99.7 | — | Surya OCR 2 Model Card | 2026-10-24 |
The model performs OCR, layout analysis, and table recognition for document intelligence tasks.[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.
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.
Listed in LLMs
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | Qwen3_5ForConditionalGeneration |
| Model type | qwen3_5 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 686,154,304 |
| Size band | S |
| MoE | no |
| Context length | 262,144 (config.json max_position_embeddings) |
| Papers | arXiv:2105.15203 |
| License | OpenRAIL (openrail) |
| Hub snapshot | 2026-10-25 |
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.
Weights-only memory for surya-ocr-2: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 1.6 GB · Edge (<4 GB) |
|---|---|
| FP8 (1 byte/param) | 0.8 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 0.4 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
finetune: 7quantized: 11adapter: 1
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| Reza2kn/Bina-0.1-Koochik | finetune | — | 666M | S | openrail | Reza2kn | 32 | 7k |
| Reza2kn/surya-ocr-2-persian-handwriting-369k | finetune | — | 666M | S | openrail | Reza2kn | 0 | 1.2k |
| aglaia-models/surya-ocr-2-mlx | quantized | mlx | 666M | S | openrail | aglaia-models | 3 | 925 |
| cstr/surya-det-GGUF | quantized | GGUF | — | — | other | cstr | 1 | 772 |
| Reza2kn/surya-ocr-2-nvfp4a16 | quantized | — | 531M | S | openrail | Reza2kn | 2 | 411 |
| Reza2kn/surya-ocr-2-mlx-8bit-g64 | quantized | mlx | 666M | S | openrail | Reza2kn | 0 | 272 |
| Reza2kn/surya-ocr-2-mlx-4bit-g64 | quantized | mlx | 666M | S | openrail | Reza2kn | 0 | 236 |
| Remidesbois/surya-bubble-ocr-poneglyph | quantized | — | 666M | S | — | Remidesbois | 0 | 229 |
| Reza2kn/Bina-0.1-Koochik-FP16 | quantized | -FP16 | 666M | S | openrail | Reza2kn | 0 | 121 |
| datalab-to/surya-parsebench-rl | finetune | — | 666M | S | openrail | datalab-to | 1 | 23 |
| gojiberries/savitr | quantized | — | 666M | S | other | gojiberries | 0 | 71 |
| PersianML/Bina-0.1-Koochik | finetune | — | 666M | S | openrail | PersianML | 0 | 68 |
| Remidesbois/surya-ocr-2-poneglyph-bbox | finetune | — | 666M | S | openrail | Remidesbois | 0 | 63 |
| Prince-1/Surya-Ocr-2-Onnx | quantized | — | — | — | openrail | Prince-1 | 1 | 0 |
| usermma/surya-ocr-2-abliterated | finetune | — | 565M | S | — | usermma | 0 | 44 |
| Reza2kn/surya-ocr-2-coreml-runtime | quantized | — | — | — | openrail | Reza2kn | 0 | 42 |
| gojiberries/savitr-bf16 | finetune | -bf16 | 666M | S | other | gojiberries | 0 | 35 |
| ANZ-Innovation/VietSurya-v1 | adapter | — | — | — | openrail | ANZ-Innovation | 0 | 13 |
| Reza2kn/surya-ocr-2-onnx-fp16-split-canary | quantized | -fp16 | — | — | openrail | Reza2kn | 0 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2026-05-27 | Add olmOCR-bench eval results | 3b3d4cd |
| 2026-05-27 | Update README.md | ac7f775 |
| 2026-05-27 | Update README.md | 9ec8e5c |
| 2026-05-27 | Add Datalab logo at top of README | 1a4ae4b |
| 2026-05-27 | Update README.md | a86fad4 |
| 2026-05-27 | Update README.md | 3da9109 |
| 2026-05-27 | Update README.md | 8e5c16a |
| 2026-05-27 | Move 28 demo images into assets/ subfolder | 8391187 |
| 2026-05-26 | Refresh olmOCR-bench chart (cleaner styling + pareto frontier) | 27b63e8 |
| 2026-05-26 | Update README.md | 2c267e5 |
| 2026-05-26 | Sync README + screenshots + chart; bump license threshold to $5M | 41f0836 |
| 2026-05-26 | Update weights from datalab-to/surya-2.1.6 (dedup tied lm_head) | 053b963 |
| 2026-05-17 | Update throughput numbers (5090 conc-sweep + Mac single row) | 14bf041 |
| 2026-05-14 | Add README + example/chart images | 20716e4 |
| 2026-05-14 | Initial: dedup'd weights from datalab-to/surya-2.1.2-mtp + LICENSE from datalab-to/chandra-ocr-2 | ee38d9b |
| 2026-05-14 | initial commit | 2af02e2 |
Hosted endpoints observed for surya-ocr-2, 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 | datalab-to/surya-ocr-2 | — | — | — | — | — | — | — | live | 2026-10-25 |
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.
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).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:38 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
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
