Language modelActive
Solar Mini 4
Upstage
- Released
- August 11, 2026
- Data date
- October 3, 2026
Solar Mini 4 combines a 512K context window with an MoE model that activates just 3 of its 35 billion parameters per token. Upstage targets repeated agent calls at high volume. For local deployment, the provider specifies a single H100 with 80 GB of memory when using quantized weights.
The model can generate up to 128K output tokens. Upstage offers it through the Console API, Solar Chat, and OpenRouter, as well as for on-premises deployment.
Specifications and access
| Specification | Value and source |
|---|---|
| Model class | Mixture-of-experts language model for agentsSource |
| Parameters | 35 billion total, 3 billion activeSource |
| Context window | 512K tokens according to UpstageSource |
| Maximum output | 128K tokens according to UpstageSource |
| Model ID | solar-mini4Source |
| Access | Upstage Console API, Solar Chat, and OpenRouterSource |
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
AA-LCR (Show measurement, test conditions, and source)
- Source value
- 83.3
- Score
- 83.3
- Metric
- AA-LCR
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- upstage.ai
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Upstage vendor table; hardware throughput is an internal 2xH100 measurement and not a benchmark.
- Context
- Source-specific observation; it is not a shared comparison cohort.
SciCode (Show measurement, test conditions, and source)
- Source value
- 47.6
- Score
- 47.6
- Metric
- SciCode
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- upstage.ai
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Upstage vendor table; hardware throughput is an internal 2xH100 measurement and not a benchmark.
- Context
- Source-specific observation; it is not a shared comparison cohort.
HLE (Show measurement, test conditions, and source)
- Source value
- 25.8
- Score
- 25.8
- Metric
- HLE
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- upstage.ai
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Upstage vendor table; hardware throughput is an internal 2xH100 measurement and not a benchmark.
- Context
- Source-specific observation; it is not a shared comparison cohort.
AutomationBench-AA (Show measurement, test conditions, and source)
- Source value
- 22.3
- Score
- 22.3
- Metric
- AutomationBench-AA
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- upstage.ai
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Upstage vendor table; hardware throughput is an internal 2xH100 measurement and not a benchmark.
- Context
- Source-specific observation; it is not a shared comparison cohort.
Artificial Analysis Intelligence Index v4.3.2 (Show measurement, test conditions, and source)
- Source value
- 24.1
- Score
- 24.1
- Metric
- Artificial Analysis Intelligence Index v4.3.2 reported score
- Unit
- No unit provided
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- upstage.ai
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Upstage vendor table; hardware throughput is an internal 2xH100 measurement and not a benchmark.
- Context
- Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.
tau3 Banking (Show measurement, test conditions, and source)
- Source value
- 47.2
- Score
- 47.2
- Metric
- tau3 Banking reported score
- Unit
- No unit provided
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- upstage.ai
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Upstage vendor table; hardware throughput is an internal 2xH100 measurement and not a benchmark.
- Context
- Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.
Sources and data date
Every statement links to its underlying documentation or leaderboard.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 4, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Additional source | Upstage Solar Mini 4 announcement (retrieved October 3, 2026; October 4, 2026) · Upstage Solar Mini 4 announcement · Editorial description reviewed October 4, 2026 |