On 22 July 2026, Upstage announced Solar Open 2 250B-A15B. Upstage scaled its Korean-first Solar line into a sparse long-context agent model trained for tools, office work and coding.
This release brief was checked against first-party material on 10 August 2026. The date above is the public announcement date, not the date a repository was created or a third-party provider added the model. Where access or weights arrived later, that distinction is recorded below.
Release record
| Field | Verified detail |
|---|---|
| Announcement | 22 July 2026 |
| Availability or weight release | 22 July 2026 as weights, technical report, playground and serving recipes |
| Release type | open-weight multilingual agent foundation model |
| Access | downloadable weights under the Upstage Solar Licence with commercial and derivative-model conditions |
| Architecture | a 250B-total, 15B-active hybrid-attention MoE with 320 routed experts, one shared expert and NoPE |
| Maximum stated context | one million tokens |
What changed
Solar Open 2 combines a very large total parameter count with only 15B active per token and keeps KV cache on 12 of 48 layers. The model officially supports Korean, English and Japanese and publishes both weights and a detailed report. Its custom licence permits commercial use but imposes naming, attribution and notice conditions on derivative models, so “open weights” does not mean Apache-style freedom.
The practical comparison is therefore not simply whether Solar Open 2 250B-A15B has the largest headline score. Teams need to ask whether its architecture, access terms, latency, tool behaviour and evaluation setup match the workload they actually intend to run. A model can lead one harness while losing on cost, refusal behaviour, multilingual quality or repeatability in another.
Benchmarks worth retaining
| Evaluation | Reported result | How to read it |
|---|---|---|
| LiveCodeBench | 92.4 | Highest reported score in Upstage’s comparison group |
| MMLU-Pro | 86.2 | Release-time knowledge and reasoning result |
| Ko-GDPval | 86.8 | Korean workplace-agent score across 170 scenarios and 58 professions |
These are release-time results, not independently reproduced guarantees. Upstage’s leading comparisons use its disclosed harnesses and model settings, and the one-million-token maximum needs task-level retrieval and needle tests before production claims. Scores should remain attached to the disclosed effort setting, agent harness, tool access, timeout, context-management policy and judge model. Moving a number into a procurement sheet without those conditions creates false comparability.
Architecture and access
Solar Open 2 250B-A15B is described as a 250B-total, 15B-active hybrid-attention MoE with 320 routed experts, one shared expert and NoPE with one million tokens of stated context. Its access position at verification time is downloadable weights under the Upstage Solar Licence with commercial and derivative-model conditions. That wording matters: open weights, source-available weights, an API, a product preview and a research demonstration give adopters very different rights and different levels of reproducibility.
Before deployment, record the exact model identifier or checkpoint, inference stack, quantisation, reasoning setting, region, price schedule and supplier terms. If the release uses a custom licence, read the licence itself rather than relying on the word “open” in launch copy. If it is API-only, preserve the dated documentation and change-notice route because the served snapshot can change without a downloadable artefact.
What an evaluation should test next
For Solar Open 2 250B-A15B, a credible internal gate should include:
- a frozen set of representative tasks with pass, fail and abstain criteria;
- a matched baseline using the same tools, timeout, prompt budget and reviewer rubric;
- repeated runs to expose variance rather than reporting a single best attempt;
- latency, token use and total task cost alongside task success;
- adversarial, multilingual and long-context cases relevant to the real deployment; and
- rollback evidence showing the previous model can be restored safely.
The wider model change-control guide explains how to keep model, prompt, tool and corpus changes reconstructable. The AI dependency inventory guide covers the release and supplier records needed after deployment.
AIEngine verdict
Solar Open 2 is a major sovereign-AI and open-weight release. Its 15B-active footprint is attractive, but the infrastructure requirement and custom derivative terms must be included in total-cost and legal review.
This is a launch assessment, not a certification. Benchmark leadership is useful evidence of where to test; it is not authorization to place the model in a high-impact workflow without domain evaluation, security review and an accountable owner.
Primary sources
Image provenance
Hero image: Upstage official release artwork. The locally served WebP is a crop of the first-party release or model-card asset recorded in the repository provenance manifest.
