Upstage

Solar Mini 4

Upstage

Solar Mini 4 is Upstage’s compact agent model for high-volume retrieval, structured generation, and tool-driven workflows. Its mixture-of-experts architecture has 35B total parameters and 3B active per token, with a 512K context and up to 128K output tokens. It supports parallel function calling and is offered through Upstage’s hosted API, with on-premises deployment options. The model emphasizes compute efficiency for repetitive work while retaining long-context reasoning. On-premises availability does not imply publicly downloadable weights; memory requirements still reflect the full 35B parameters.

Solar Pro 4

Upstage

Solar Pro 4 is Upstage’s hosted commercial flagship for long-running enterprise agents. It focuses on terminal tasks, multi-turn tool use, document-grounded reasoning, and producing deliverables across multiple files. It supports English, Korean, and Japanese, adjustable reasoning effort, a 512K context, and up to 128K output tokens. The model is offered through Upstage Console and partner APIs rather than open weights. Its August 11 launch blog precedes the August 20 press release. Benchmarks depend on agent environments and do not guarantee task completion in a particular production workflow.

Solar Open 2

Upstage

Solar Open 2 is Upstage’s 250B-parameter open-weight model for document-heavy agents, office productivity, and coding. It activates 15B parameters per token and interleaves linear attention with standard attention to improve long-context inference efficiency. A 1M-token context supports large document collections and extended tool histories. Its release includes weights and a technical report describing training and evaluation. The custom Solar Open license governs redistribution and commercial use. Sparse activation lowers computation relative to a dense model of the same total size but does not eliminate the storage and memory requirements of the full weights.

Solar Pro 3

Upstage

Solar Pro 3 is Upstage’s hosted enterprise language model focused on reasoning, instruction following, and Korean and English workflows. Its 102B sparse architecture activates 12B parameters. The January launch established public API access; a March update strengthened agentic tasks through SnapPO reinforcement learning while preserving the service interface. The two documented snapshots are grouped under this model identity. API and partner access support enterprise assistants, but the Pro checkpoint is not a public weight release.

Solar Open 100B

Upstage

Solar Open 100B is Upstage’s first large open-weight Solar mixture-of-experts model, with 102.6B total parameters and 12B active per token. It targets English and Korean reasoning, instruction following, coding, and agentic work with 128K context. Although a model card existed earlier in December, weight-bearing repository commits appeared on December 31; the earlier teaser is not treated as availability. Current weights use the custom Upstage Solar License while accompanying code uses Apache 2.0. This is the first-generation Solar Open checkpoint, distinct from Solar Open 2’s later 250B hybrid-attention architecture.

Solar Pro 2

Upstage

Solar Pro 2 is Upstage’s 31B hosted language model for reasoning, multilingual assistance, and tool-driven enterprise workflows. Its public Preview opened on May 20, 2025 with 64K context, followed by the commercial launch on July 10. Users can select faster chat responses or a reasoning mode for more involved tasks; these operating modes do not imply a disclosed hybrid network architecture. The generation emphasizes English, Korean, and Japanese and is offered through Upstage Console APIs and enterprise deployment arrangements. Public model weights are not published with this release.

Solar Pro

Upstage

Solar Pro is Upstage’s 22B dense language model generation, introduced with downloadable MIT-licensed Preview weights and hosted access in September 2024. The initial Preview is English-only with a 4K context and no system-prompt support. The commercial release announced in December extends the generation to English and Korean and supports deployment through Upstage and AWS managed endpoints. This family page preserves the distinction between the openly downloadable Preview and the separately licensed commercial serving model; downloadable Preview weights do not establish identical weights or MIT rights for every hosted revision.

SOLAR 10.7B

Upstage

SOLAR 10.7B is Upstage’s original depth-up-scaled language model, built by expanding a Mistral 7B initialization into a 48-layer dense transformer and continuing pretraining. The release includes a base checkpoint and an instruction-tuned conversational variant, also marketed as Solar Mini. The base weights use Apache 2.0; the current instruction checkpoint specifies CC BY-NC 4.0 because its alignment data includes non-commercial datasets. The documented native context is 4,096 tokens. It supports local text generation, customization, and research without assuming that later hosted Solar Mini revisions are the identical checkpoint.