Upstage Solar Mini 4: Intelligence Index 24 at $0.10 Per Million Tokens, 2x the Median for Its Price Tier
Upstage released Solar Mini 4 on September 22 at $0.10 per million input tokens and $0.40 per million output tokens, with a 90% cache discount. Artificial Analysis puts its Intelligence Index at 24, against a median of 12 for reasoning models in its price tier — twice the typical performance at that cost point.
The Numbers
| Metric | Value |
|---|---|
| AA Intelligence Index | 24 |
| Tier median (reasoning models) | 12 |
| Input cost | $0.10 / 1M tokens |
| Output cost | $0.40 / 1M tokens |
| Cache discount | 90% |
| Speed tier | Top (4/4) |
| Knowledge cutoff | February 2026 |
| API providers | 1 |
Solar Mini 4 sits alongside Solar Pro 4 — Upstage’s August release at $0.30/$1.20 per million tokens. The Mini serves a distinct use case: pipelines where per-token cost is the binding constraint and a model in the mid-twenties on Intelligence Index is sufficient for the task.
Where the Math Works
At $0.10 input, a 10-million-token daily workload costs $1. The same workload at GPT-6.1 Sol’s $2 input rate costs $20. For document classification, customer query routing, or internal automation that operates at scale, Solar Mini 4’s efficiency profile is the argument — not its raw capability ceiling.
The 90% cache discount amplifies that. Applications with repetitive system prompts or fixed context windows pay 10% of the input rate on cache hits. At that level, token cost effectively disappears as a concern for cached portions of context.
The knowledge cutoff of February 2026 is current enough for most production tasks. Applications that require awareness of developments from the past eight months will need retrieval augmentation.
Upstage’s Position
Upstage built its business on domain-specific fine-tuning and document AI. Solar Pro 4 in August was its entry into general-capability leaderboard competition. Solar Mini 4 extends downmarket into the segment where Korean enterprise customers, cost-constrained startups, and high-volume API users operate.
One API provider currently serves Solar Mini 4. That single-provider dependency carries operational risk for production systems — provider downtime has no fallback path. Models like Llama and Qwen with multi-provider distribution offer redundancy that Solar Mini 4 does not.
The efficiency case is real. The infrastructure risk is also real. The two considerations define the decision for any team evaluating it.