This page provides our model catalog, showcasing a range of AI models designed to enhance automation and intelligence in your workflows. Whether you need Solar LLM for advanced language processing or specialized models for document parsing, this catalog helps you explore their capabilities and deployment options. For detailed usage instructions, see the Capabilities menu.
Refer to the Available platforms descriptions to see which platforms offer immediate self-serve access for each model.
All models listed on this page can also be deployed on a virtual private cloud (VPC) with any cloud provider or on-premises, unless otherwise noted.
Contact us for more details on these options.
Solar LLM
solar-open2 (Deprecated)
solar-open2 was offered as a closed beta to a limited group of users. API serving ended on August 12, 2026, and the model is no longer available through the Upstage API. It remains available as open weights on Hugging Face.
Solar Open 2 is Upstage's open-weight model, optimized for agentic use with long-context support and full Korean, Japanese, and English coverage.
| Release date | 2026-07-08 |
| Parameters | Undisclosed |
| Context length | Undisclosed |
| Training data cut-off | Feb 2026 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, reasoning, structured outputs, function calling |
| Available platforms |
|
solar-pro4-260806
Solar Pro 4 is Upstage's flagship model, specialized for agentic use. It is particularly strong at document-based work, coding, and business-related tasks, and substantially improves agent performance compared to Solar Pro 3.
| Release date | 2026-08-06 |
| Parameters | Undisclosed |
| Context length | 512K |
| Max output tokens | 128K |
| Training data cut-off | Feb 2026 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, reasoning, structured outputs, tool calling |
| Available platforms |
|
solar-pro3-260323 (Deprecated)
This version will be fully deprecated on October 30, 2026 (KST). Requests naming it will return an error after that date. Please migrate to the solar-pro4 alias.
This release includes the following updates.
- Improved Agentic Performance.
- Improved Deep Reasoning performance.
solar-pro3-260126 (Deprecated)
This version was deprecated on April 30, 2026. Please upgrade to the solar-pro4 alias.
Solar Pro 3 is Upstage's powerful Mixture-of-Experts (MoE) language model. With 102B total parameters and 12B active parameters per forward pass, it delivers exceptional performance while maintaining computational efficiency. The model features a 128K context length and is optimized for Korean language tasks while supporting English and Japanese.
| Release date | 2026-01-26 |
| Parameters | 102B total / 12B active (MoE) |
| Context length | 128,000 |
| Training data cut-off | Jul 2025 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, reasoning, structured outputs, function calling, parallel tool calls |
| Available platforms |
solar-pro2-251215 (Deprecated)
This version will be fully deprecated on October 30, 2026 (KST). Requests naming it will return an error after that date. Please migrate to the solar-pro4 alias.
This release includes the following updates.
- Improved inference efficiency through optimized quantization.
- Fixed an issue where parallel tool calls could result in incorrect streaming chunk ordering.
- Resolved an issue where reasoning was not properly triggered when using guided decoding.
- Fixed response_format handling to work correctly with reasoning mode.
- Added support for NVIDIA Blackwell and ARM64 architectures.
| Release date | 2025-12-15 |
| Parameters | 31B |
| Context length | 65,536 |
| Training data cut-off | Mar 2025 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, reasoning, structured outputs, function calling |
| Available platforms |
|
solar-pro2-250909 (Deprecated)
This version was deprecated on December 29, 2025. Please upgrade to the solar-pro4 alias.
This release includes the following updates.
- Resolved an issue where the model's response could occasionally repeat itself when processing tool calls.
| Release date | 2025-09-09 |
| Parameters | 31B |
| Context length | 65,536 |
| Training data cut-off | Mar 2025 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, reasoning, structured outputs, function calling |
| Available platforms |
solar-pro2-250710 (Deprecated)
This version was deprecated on October 14, 2025. Please upgrade to the solar-pro4 alias.
The first official release of Solar Pro 2. Solar Pro 2 is Upstageβs flagship model, recognized as a 2025 frontier model by Artificial Analysis. It achieved an Intelligence Index score of 58, demonstrating strong performance across reasoning, math, coding, and real-world knowledge tasks. With 31B parameters, it delivers frontier-level accuracy and efficiency, optimized for enterprise use. Learn more.
| Release date | 2025-07-10 |
| Parameters | 31B |
| Context length | 65,536 |
| Training data cut-off | Mar 2025 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, reasoning, structured outputs, function calling |
| Available platforms |
solar-pro2-preview (Deprecated)
This version was deprecated on July 15, 2025. Please upgrade to the solar-pro4 alias.
A high-performance model from Upstage that supports both Chat and Reasoning modes, allowing users to tailor performance based on task complexity. Reasoning mode provides structured, multi-step thinking for complex tasks, while Chat mode delivers fast, direct responses for simpler queries. By supporting both modes in a single model, this design achieves an optimal balance of speed, accuracy, and resource efficiency for diverse use cases. Learn more.
| Release date | 2025-05-20 |
| Parameters | 31B |
| Context length | 65,536 |
| Training data cut-off | Mar 2025 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, reasoning (Experimental support: Structured outputs, function calling) |
solar-pro-250422 (Deprecated)
This version was deprecated on August 13, 2025. Please upgrade to the solar-pro4 alias.
A high-performance upgrade with stronger Korean fluency for enterprise-grade tasks. Learn more.
| Parameters | 22B |
| Context length | 32,768 |
| Training data cut-off | Sep 2024 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, structured outputs, function calling |
| Available platforms |
|
solar-pro-241126 (Deprecated)
This version has been deprecated. Please upgrade to the solar-pro4 alias.
An enterprise-grade LLM designed for exceptional instruction-following and processing structured formats like HTML and Markdown. It delivers top multilingual performance in English, Korean, and Japanese, with domain expertise in Finance, Healthcare, and Legal. Learn more.
| Parameters | 22B |
| Context length | 32,768 |
| Training data cut-off | May 2024 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, structured outputs, function calling |
solar-pro-preview
This is an older model. For better performance, consider using recent versions of Solar Mini or Solar Pro.
The initial, open-source version of Solar Pro. This large language model delivers the performance of a 70B+ parameter model while running efficiently on a single GPU. Learn more.
| Release date | 2024-09-10 |
| Parameters | 22B |
| Context length | 32,768 |
| Training data cut-off | May 2024 |
| Supported languages | English |
| Capabilities | Chat |
| Available platforms |
solar-mini4-260922
Solar Mini 4 is Upstage's cost-efficient, compact language model, built for agentic use cases where response speed and cost matter. It combines fluent Korean with strong English and Japanese capabilities and supports long-context workloads.
| Release date | 2026-09-22 |
| Parameters | 35B / 3B active |
| Context length | 512K |
| Max output tokens | 128K |
| Training data cut-off | Feb 2026 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, reasoning, structured outputs, tool calling |
| Available platforms |
|
solar-mini-250422 (Deprecated)
This version will be fully deprecated on October 30, 2026 (KST). Requests naming it will return an error after that date. Please migrate to the solar-mini4 alias.
A lightweight model thatβs now more fluent in Korean.
| Parameters | 10.7B |
| Context length | 32,768 |
| Training data cut-off | Sep 2024 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, structured outputs, function calling |
| Available platforms |
|
solar-mini-250123 (Deprecated)
This version has been deprecated. Please upgrade to the solar-mini4 alias.
A compact yet powerful 10.7-billion-parameter large language model (LLM) for businesses seeking AI solutions. Solar Mini is an instruction-following conversational model, supporting English, Korean, and Japanese. It excels at fine-tuning, providing seamless integration and high-quality language processing. Learn more.
| Parameters | 10.7B |
| Context length | 32,768 |
| Training data cut-off | Nov 1st, 2024 |
| Supported languages | English, Korean, and Japanese |
| Capabilities | Chat, structured outputs, function calling |
solar-mini-240612 (Deprecated)
This version has been deprecated. Please upgrade to the solar-mini4 alias.
| Parameters | 10.7B |
| Context length | 32,768 |
| Training data cut-off | Dec 2023 |
| Supported languages | English, Korean |
| Capabilities | Chat |
| Available platforms |
solar-mini-ja-250123 (Deprecated)
This version was deprecated on April 10, 2025. Please upgrade to the solar-mini4 alias.
Includes function calling and updated performance.
| Parameters | 10.7B |
| Context length | 32,768 |
| Training data cut-off | Dec 2023 |
| Supported languages | Japanese |
| Capabilities | Chat, function calling |
solar-mini-ja-240612 (Deprecated)
This version was deprecated on April 10, 2025. Please upgrade to the solar-mini4 alias.
Improved text detection for single characters and special characters. Learn more.
| Parameters | 10.7B |
| Context length | 32,768 |
| Training data cut-off | Dec 2023 |
| Supported languages | Japanese |
| Capabilities | Chat |
| Available platforms |
solar-10.7b-v1.0
This is an older model. For better performance, consider using recent versions of Solar Mini or Solar Pro.
The initial, open-source version of Solar Mini. Learn more.
| Release date | 2024-01-25 |
| Parameters | 10.7B |
| Supported languages | English |
| Capabilities | Chat |
| Available platforms |
solar-docvision-preview
An experimental vision LLM specialized for document visual question answering (VQA).
| Release date | 2024-09-10 |
| Supported languages | English |
| Context length | 8192 |
| Capabilities | Document VQA |
| Available platforms |
syn-250326 (Deprecated)
This version will be fully deprecated on October 30, 2026 (KST). Requests naming it will return an error after that date. Please migrate to the solar-mini4 alias.
A compact and reliable 10B-class Japanese LLM co-developed by Upstage and Karakuri. Syn delivers strong instruction-following and structured format processing, with domain strength in Finance, Healthcare, and Legal. Designed for efficient fine-tuning and scalable enterprise deployment. Learn more.
| Parameters | 10.7B |
| Context length | 32,768 |
| Training data cut-off | Jan 2024 |
| Supported languages | Japanese, English |
| Capabilities | Chat |
| Available platforms |
syn-pro-251021
Syn Pro is an advanced Japanese LLM co-developed by Upstage and Karakuri, featuring approximately 30 billion parametersβtriple its predecessor Syn. Through sophisticated fine-tuning techniques and advanced Japanese dataset utilization, Syn Pro deeply understands linguistic nuances, industry-specific terminology, and cultural contexts, enabling it to deliver accurate and natural responses aligned with business processes and enterprise requirements. This makes it an ideal solution for companies seeking high-performance, cost-effective AI that seamlessly integrates into Japanese business environments while maintaining computational efficiency for practical enterprise deployment.
| Parameters | 31B |
| Context length | 65,536 |
| Training data cut-off | Nov 2023 |
| Supported languages | Japanese, English and Korean |
| Capabilities | Chat |
| Available platforms |
|
Document Parse
document-parse-260930
This release includes the following updates.
- On
document-parse-260930,enhancedmode now uses an end-to-end vision-language model (VLM) parser in place of the previous multi-stage refinement pipeline, andautomode sends complex pages to this parser. Thedocument-parsealias now points todocument-parse-260930, soenhancedmode on the alias uses this parser. - For PDFs with a text layer (digital-born documents), text that the VLM parser recognizes is cross-checked against that text layer to reduce misrecognized characters.
- Richer HTML and Markdown output in
enhancedmode: heading levels (<h2>to<h6>, with the depth insub_category),<ul>/<ol>lists, code blocks with their language, inline text styling, checkboxes, and line breaks within paragraphs. - Figure and chart captions in
enhancedmode are written in the language of the document's intended readers, which the model detects automatically. Korean, English, and Japanese are currently supported. - Inline equations within text are recognized in
enhancedmode and returned as LaTeX wrapped in$...$. - Improved recognition accuracy for block equations (the
equationcategory) inenhancedmode. - Improved table accuracy and stability on dense pages in
enhancedmode.
| Release date | 2026-09-30 |
| Capabilities | Document parsing |
| Available platforms |
|
document-parse-260630 (Deprecated)
This version will be fully deprecated on November 2, 2026 (KST). Requests naming it will return an error after that date. Please migrate to the document-parse alias.
This release includes the following updates.
- Added spreadsheet (xlsx) input support for accurate HTML table output.
- Improved throughput by approximately 60%.
- Added a
wordsoption to return word-level OCR results with bounding-box coordinates. - Improved Japanese kanji and handwriting recognition with an upgraded OCR detector architecture (Stable DINO).
- Improved document orientation handling, including a fix for a 180Β° flip detection issue.
- Improved stability when processing very dense or large documents.
| Release date | 2026-06-30 |
| Capabilities | Document parsing |
| Available platforms |
|
document-parse-260128 (Deprecated)
This version will be fully deprecated on October 12, 2026 (KST). Requests naming it will return an error after that date. Please migrate to the document-parse alias.
This release includes the following updates.
- Improved Japanese handwriting recognition (CLEval: 87.04 β 88.43).
- Improved table recognition
- (Standard mode) Accuracy for tables with narrow row spacing (TEDS: 73.9 β 95.6).
- (Enhanced mode) Header cell, in-cell newline, and checkbox recognition.
- (Enhanced mode) Accuracy for nested tables (TEDS: 56.6 β 93.3).
| Release date | 2026-01-28 |
| Capabilities | Document parsing |
| Available platforms |
|
document-parse-251217 (Deprecated)
This version was deprecated on July 31, 2026. Please upgrade to the document-parse alias.
This release includes the following updates.
- Improved Japanese insurance document layout recognition (NID: 89.5 β 91.6).
- Improved Japanese insurance document table recognition (TEDS: 79.0 β 89.8).
- Improved Japanese handwriting recognition (CLEval: 86.7 β 87.0).
- Improved layout recognition (LED: 71.6 β 76.4).
| Release date | 2025-12-17 |
| Capabilities | Document parsing |
| Available platforms |
document-parse-250618 (Deprecated)
This version was deprecated on February 26, 2026. Please upgrade to the document-parse alias.
This release includes the following updates.
- Improved form document layout recognition accuracy from 96.7 to 97.0.
- Enhanced Japanese document layout recognition accuracy from 92.7 to 94.9, and table recognition accuracy from 61.8 to 76.5.
- Document rotation correction is applied by default for all documents.
| Release date | 2025-06-18 |
| Capabilities | Document parsing |
| Available platforms |
document-parse-250508 (Deprecated)
This version was deprecated on September 5, 2025. Please upgrade to the document-parse alias.
This release includes the following updates.
- Table recognition accuracy for form documents improved from 67.2 to 91.7 in our internal tests.
- Added support for nested tables and images in tables.
- Enhanced OCR accuracy for Japanese and Chinese characters, with internal test results showing accuracy improvement from 91.7 to 93.2.
| Release date | 2025-05-08 |
| Capabilities | Document parsing |
| Available platforms |
document-parse-250404 (Deprecated)
This version was deprecated on September 5, 2025. Please upgrade to the document-parse alias.
This release includes the following updates. Learn more.
- Improved performance for form application style documents.
- Enhanced handling of rotated (non-upright) documents.
- Merged tables spanning multiple pages are now supported.
- Supports long images Learn more.
- Automatically applies OCR when PDF recognition fails with
ocr="auto". - Supports
.hwp,.hwpxfile extension.
| Release date | 2025-04-04 |
| Capabilities | Document parsing |
| Available platforms |
document-parse-250116 (Deprecated)
This version was deprecated on May 15, 2025. Please upgrade to the document-parse alias.
The chart_recognition feature has been added and the inference speed has improved.
Learn more.
| Release date | 2025-01-16 |
| Capabilities | Document parsing |
document-parse-240910 (Deprecated)
This version has been deprecated. Please upgrade to the document-parse alias.
Added support for Microsoft Word, Excel, and Powerpoint. Markdown output for tables and list items. Base64 encoding of extracted images for all requested layout categories. Learn more.
| Release date | 2024-09-10 |
| Capabilities | Document parsing |
layout-analysis-0.4.0 beta
Improved the accuracy for table recognition. Added new layout elements: heading1, list, index, and footnote. Changed the default value for ocr field to false.
| Release date | 2024-07-04 |
| Capabilities | Document parsing |
layout-analysis-0.3.1 beta (Deprecated)
This version has been deprecated. Please upgrade to the document-parse alias.
Fixed a bug where extracted text from table elements was truncated.
| Release date | 2024-06-17 |
| Capabilities | Document parsing |
layout-analysis-0.3.0 beta (Deprecated)
This version has been deprecated. Please upgrade to the document-parse alias.
Improved the inference speed by 2x for digital-born PDF documents.
| Release date | 2024-06-11 |
| Capabilities | Document parsing |
layout-analysis-0.2.1 beta (Deprecated)
This version has been deprecated. Please upgrade to the document-parse alias.
Removed unnecessary <thead> tags from table elements and fixed bugs.
| Release date | 2024-05-02 |
| Capabilities | Document parsing |
layout-analysis-0.2.0 beta (Deprecated)
This version has been deprecated. Please upgrade to the document-parse alias.
Improved the accuracy for table recognition and performance for layout detection.
| Release date | 2024-04-04 |
| Capabilities | Document parsing |
layout-analyzer-0.1.0 beta (Deprecated)
This version has been deprecated. Please upgrade to the document-parse alias.
A layout analyzer model which detects elements within a document, recognizes tables, and serializes elements according to reading order.
| Release date | 2024-02-28 |
| Capabilities | Document parsing |
Information Extract
information-extract-260904
| Release date | 2026-09-04 |
| Capabilities | Information extraction |
| Available platforms |
|
information-extract-260610 (Deprecated)
This version will be fully deprecated on October 6, 2026 (KST). Requests naming it will return an error after that date. Please migrate to the information-extract alias.
This release includes the following updates.
- Improved extract performance
- Improved confidence performance
| Release date | 2026-06-10 |
| Capabilities | Information extraction |
| Available platforms |
|
information-extract-260304 (Deprecated)
This version will be fully deprecated on October 6, 2026 (KST). Requests naming it will return an error after that date. Please migrate to the information-extract alias.
This release includes the following updates.
- Improved Japanese handwriting recognition
- Improved table recognition
- Schema generation is improved to generate more group schemas for documents with tables.
- Location coordinates with
alllevel granularity action is changed to return fallback coordinates.
| Release date | 2026-03-04 |
| Capabilities | Information extraction |
| Available platforms |
|
information-extract-260114 (Deprecated)
This version was deprecated on June 16, 2026. Please upgrade to the information-extract alias.
Enhanced extrction for japanese documents and table extraction. Word level location coordinates feature is added.
| Release date | 2026-01-14 |
| Capabilities | Information extraction |
| Available platforms |
information-extract-250930 (Deprecated)
This version was deprecated on February 18, 2026. Please upgrade to the information-extract alias.
Enhanced table extraction capabilities, especially for large tables.
| Release date | 2025-09-30 |
| Capabilities | Information extraction |
| Available platforms |
information-extract-250804 beta (Deprecated)
This version was deprecated on February 18, 2026. Please upgrade to the information-extract alias.
Improved performance for form application style documents and rotated documents.
| Release date | 2025-08-04 |
| Capabilities | Information extraction |
| Available platforms |
information-extract-250630 beta (Deprecated)
This version was deprecated on September 5, 2025. Please upgrade to the information-extract alias.
Fixed an intermittent bug that caused duplicate information extraction.
| Release date | 2025-06-30 |
| Capabilities | Information extraction |
information-extract-250529 beta (Deprecated)
This version was deprecated on September 5, 2025. Please upgrade to the information-extract alias.
First release of Upstage Information Extract. Information Extract enables high-accuracy key-value extraction from any document type with zero training. Learn more.
| Release date | 2025-05-29 |
| Capabilities | Information extraction |
Document classification
document-classify-260304
This release includes the following updates.
- Improved Japanese handwriting recognition
- Improved table recognition
| Release date | 2026-03-04 |
| Capabilities | Document classification |
| Available platforms |
|
document-classify-260114
Enhanced classification for japanese documents. Document split feature is added.
| Release date | 2026-01-14 |
| Capabilities | Document classification |
| Available platforms |
|
document-classify-250830 beta (Deprecated)
This version was deprecated on February 26, 2026. Please upgrade to the document-classify alias.
Initial release
| Release date | 2025-08-30 |
| Capabilities | Document classification |
| Available platforms |
Prebuilt information extraction models
These models were deprecated on May 6, 2026.
air-waybill-extraction-250415 (Deprecated)
Improved extraction accuracy from 79% to 96%, along with enhanced visual grounding accuracy.
| Field descriptions | Same as in air-waybill-extraction-4.1.6 |
air-waybill-extraction-4.1.6 (Deprecated)
An extractor model for air waybill (AWB).
| Release date | 2024-06-11 |
bill-of-lading-and-shipping-request-extraction-250415 (Deprecated)
Improved extraction accuracy from 69% to 90% and visual grounding accuracy.
bill-of-lading-and-shipping-request-extraction-4.1.6 (Deprecated)
An consolidated extractor model for bill of lading (BL or BoL) and shipping request (SR).
| Release date | 2024-06-11 |
commercial-invoice-and-packing-list-extraction-250415 (Deprecated)
Improved extraction accuracy from 74% to 84% and visual grounding accuracy.
commercial-invoice-and-packing-list-extraction-4.1.6 (Deprecated)
An consolidated extractor model for commercial invoice (CI) and packing list (PL).
| Release date | 2024-06-11 |
kr-export-declaration-certificate-extraction-250415 (Deprecated)
Improved extraction accuracy from 98% to 99% and visual grounding accuracy.
| Field descriptions | Same as in kr-export-declaration-certificate-extraction-4.1.6 |
kr-export-declaration-certificate-extraction-4.1.6 (Deprecated)
An extractor model for the export declaration certificate of Korea.
| Release date | 2024-06-11 |
receipt-extraction-3.2.0 (Deprecated)
An extractor model for paper receipts, that include store descriptions and list of items. Works best for Korean and English receipts. Improved accuracy and performance.
| Release date | 2024-04-11 |
| Available platforms | Upstage Console |
receipt-extractor-1.0.0 (Deprecated)
An extractor model for paper receipts, that include store descriptions and list of items. Works best for Korean receipts.
| Release date | 2023-04-11 |
Document OCR
ocr-260930
Document OCR now runs on the same engine as document-parse-260930. The ocr alias now points to ocr-260930.
| Release date | 2026-09-30 |
| Capabilities | Document OCR |
| Available platforms |
|
ocr-260630 (Deprecated)
This version will be fully deprecated on November 2, 2026 (KST). Requests naming it will return an error after that date. Please migrate to the ocr alias.
Strengthened Japanese support with expanded recognition of Japanese kanji and special symbols, along with improved handwriting recognition accuracy. Introduced a DETR-based text detector that improves detection of large text and words with narrow line spacing while increasing throughput by 60%. Serving stability was also improved through patchify logic and concurrent request processing enhancements.
| Release date | 2026-06-30 |
| Capabilities | Document OCR |
| Available platforms |
|
ocr-250904 (Deprecated)
This version will be fully deprecated on October 12, 2026 (KST). Requests naming it will return an error after that date. Please migrate to the ocr alias.
Improved Japanese text recognition performance with accuracy gains in both scene text (89.39% β 91.70%) and document text (95.19% β 96.72%). Maximum number of pages per file increased from 30 to 100 pages.
| Release date | 2025-09-04 |
| Capabilities | Document OCR |
| Available platforms |
|
ocr-2.2.1 (Deprecated)
This version was deprecated on October 13, 2025. Please upgrade to the ocr alias.
An OCR model specialized for English and Korean, with additional support for Japanese and Chinese character sets (Hanja, Hanzi, and Kanji). Resilient against real-world images, including wrinkled papers and rotated text.
| Release date | 2024-06-11 |
| Capabilities | Document OCR |
| Available platforms |
ocr-2.1.1 (Deprecated)
This version has been deprecated. Please upgrade to the ocr alias.
Improved text detection for single characters and special characters.
| Release date | 2024-04-04 |
| Capabilities | Document OCR |
ocr-2.1.0 (Deprecated)
This version has been deprecated. Please upgrade to the ocr alias.
Additional support for Hanja, Hanzi and Kanji. Improved accuracy and performance.
| Release date | 2024-02-28 |
| Capabilities | Document OCR |
ocr-1.0.0 (Deprecated)
This version has been deprecated. Please upgrade to the ocr alias.
An OCR model specialized for English and Korean. Resilient against real-world images, including wrinkled papers and rotated text.
| Release date | 2023-04-10 |
| Capabilities | Document OCR |
Embed
solar-embedding-2-query
Query Embedding model with an 8k context limit and 1024-dimensional vectors. Optimized for embedding user's questions in information-seeking tasks such as retrieval & reranking.
| Release date | 2026-06-30 |
| Capabilities | Embeddings |
| Context length | 8,000 |
| Vector dimensions | 1,024 |
| Available platforms |
|
solar-embedding-2-passage
Passage Embedding model with an 8k context limit and 1024-dimensional vectors. Optimized for embedding documents or texts to be searched.
| Release date | 2026-06-30 |
| Capabilities | Embeddings |
| Context length | 8,000 |
| Vector dimensions | 1,024 |
| Available platforms |
|
solar-embedding-1-large-query (Deprecated)
This version will be fully deprecated on December 31, 2026 (UTC). Requests naming it will return an error after that date. Please migrate to solar-embedding-2-query.
Solar-based Query Embedding model with a 4k context limit. This model is optimized for embedding user's question in information-seeking tasks such as retrieval & reranking. Learn more.
| Release date | 2024-05-10 |
| Capabilities | Embeddings |
| Context length | 4,000 |
| Available platforms |
|
solar-embedding-1-large-passage (Deprecated)
This version will be fully deprecated on December 31, 2026 (UTC). Requests naming it will return an error after that date. Please migrate to solar-embedding-2-passage.
Solar-based Passage Embedding model with a 4k context limit. This model is optimized for embedding documents or texts to be searched. Learn more.
| Release date | 2024-05-10 |
| Capabilities | Embeddings |
| Context length | 4,000 |
| Available platforms | Upstage Console |
Groundedness Check
groundedness-check-240502 beta (Deprecated)
This version was deprecated on October 21, 2025.
Solar Mini based groundedness checking model. (Previously called "solar-1-mini-groundedness-check") Learn more.
| Release date | 2024-05-02 |
| Context length | 32,768 |
| Capabilities | Groundedness checking |
| Available platforms | Upstage Console |
Solar Decide
solar-decide beta
Solar Decide is Upstage's structured decision model, served as a System One endpoint on Solar Mini 4. Send a state along with typed questions, and it returns a choice, a score, or a yes/no answer, each with a calibrated probability taken directly from the model rather than written out as text. Because it generates no prose, each decision takes a single forward pass and output tokens are free. With a 512K context window, an entire document can serve as the state. Solar Decide uses the same /v1/systemone schema as Jev, bringing Solar Mini 4's strong Korean understanding to routing, classification, and policy checks.
| Release date | 2026-09-22 |
| Base model | Solar Mini 4 |
| Context length | 512K |
| Endpoint | POST /v1/systemone |
| Question types | choice, score, noul |
| Capabilities | Structured decisions with calibrated probabilities |
| Available platforms | Upstage Console |