Solar Mini 4 is Upstage's new compact mixture-of-experts model, and my verdict is that it's a fast, free everyday worker rather than a frontier brain.
I score it 3.5 out of 5 overall, because the speed and the free Nous Portal window are excellent while the depth of its answers is only average.
Below I go through it feature by feature, with the official specs, every free and paid route, the setup steps, a scores table and the limits I'd want to know before trusting it.
Everything here comes from my own test on video plus Upstage's console docs, OpenRouter, Nous Portal and Artificial Analysis.
Solar Mini 4 Verdict At A Glance
Here's the short version before the detail.
| Area | My score | Verdict |
|---|---|---|
| Speed | 4.5/5 | It replied almost instantly in Hermes Agent on both a test message and a research task. |
| Price | 5/5 | It's free on Nous Portal for two weeks, and the paid price is tiny. |
| Answer quality | 3/5 | It's solid for basic agentic tasks, but it isn't frontier level. |
| Context and languages | 4/5 | It reads very long inputs and works in Korean, English and Japanese. |
| Setup | 4/5 | It takes a few clicks in Hermes, apart from the free-vs-paid trap. |
| Reliability of the free route | 3/5 | Free models get rate limited, and the window closes after two weeks. |
| Overall | 3.5/5 | It's a great free second profile, and it's not a replacement for a top model. |
What Is Solar Mini 4?
Solar Mini 4 is a language model made by Upstage, an AI lab from South Korea.
Upstage calls it a "cost-efficient compact model for agentic use", which is a fair description of how it behaved for me.
It's a mixture-of-experts model, so only part of it runs for each token it writes.
| Feature | Official detail |
|---|---|
| Parameters | 35B total, with 3B active per step |
| Context window | 512K tokens on Upstage's docs, and 524,288 tokens on OpenRouter and Nous Portal |
| Maximum output | Up to 128K tokens on Upstage |
| Languages | Korean, English and Japanese, for input and output |
| Capabilities | Chat, reasoning mode, tool calling and structured outputs |
| Release date | 22 September 2026 on Upstage's console |
| Training cut-off | February 2026 |
| Licence | Proprietary, with no public weights |
| Upstage model ID | solar-mini4, an alias for solar-mini4-260922 |
I said 500,000 tokens of context in my video, and Upstage's official figure is 512K, so I'm using the official number here.
Who Made Solar Mini 4, And Is It Free?
Upstage made it, and Upstage sells it through its console and API.
It's free in two places right now, and paid everywhere else.
| Route | Price | Score | Verdict |
|---|---|---|---|
Nous Portal free variant (upstage/solar-mini4:free) |
Free for a two-week window | 5/5 | This is the best way to use it inside Hermes Agent today. |
| Upstage Console Playground | Free, with no API key needed | 4/5 | It's perfect for a quick test, but it isn't an agent. |
| OpenRouter | $0.05 in and $0.20 out per million tokens | 4/5 | It's cheap and easy if you already use OpenRouter. |
| Upstage API | $0.10 in and $0.40 out per million tokens, with 70% off through 10 October (UTC) | 4/5 | It's the maker's own route for production. |
Nous Portal paid variant (upstage/solar-mini4) |
$0.05 in and $0.20 out per million tokens | 2/5 | It's only worth it once the free window ends. |
How I Tested Solar Mini 4 For This Verdict
I ran Solar Mini 4 inside Hermes Agent through Nous Portal, using its own Hermes Desktop profile.
I started with a one-word "test" message to check the connection and the speed.
Then I gave it a real job, which was to find AI automation news from the last seven days and summarise it.
I judged it on speed, on whether it cited sources, and on whether it told me something I didn't already know.
I didn't run it through coding or terminal tasks, so for those I'm leaning on Artificial Analysis rather than my own test.
I also didn't score it on GoldieBench, because it hasn't been benched there yet.
That makes this a hands-on first verdict rather than a lab report, and I'd rather tell you that than pretend otherwise.
Feature By Feature: How Solar Mini 4 Performed
Speed
Speed is its strongest point.
When I typed "test", it came straight back saying the session was live and that memory was off.
When I asked for AI automation news from the last seven days, it returned a full breakdown quickly.
The 3B active parameters are the reason, because less of the model has to run on every step.
Research and summaries
The news task is the best evidence I have of its quality.
It returned a breakdown, the sources it used and a "what stands out" section.
It also surfaced an Oracle launch I hadn't heard about, which is the kind of find that makes a research agent useful.
Reasoning and intelligence
Artificial Analysis scores Solar Mini 4 at 24 on its Intelligence Index.
Their write-up says that's 6 points above Qwen 3.6 35B, which has the same 3B active parameters, and 16 points above Upstage's previous flagship, Solar Pro 3.
In my video I put it as scoring above models with ten times the active parameters, which is how the launch was framed.
Even so, it's not frontier level, and it's nowhere near Claude Opus 5.5 on hard problems.
Long context
Upstage built it for long-context workloads, and a 512K window covers whole reports, transcripts or codebases in one go.
Artificial Analysis also reported a strong 83% on its long-context reasoning test, AA-LCR.
Tool calling and structured outputs
Upstage lists tool calling and structured outputs as supported, which is what an agent needs to use tools and return clean data.
That's why it works as a Hermes Agent brain at all.
Weak spots in the data
Artificial Analysis flags it as very verbose, using about 88,000 output tokens per Intelligence Index task.
It scored 1% on Terminal-Bench 4.0 in that same write-up, so I wouldn't hand it terminal-heavy coding jobs.
Its knowledge test score was weak too, with 18% accuracy on AA-Omniscience.
🔥 Want my weekly breakdown of which new models are worth your time? Inside the AI Profit Boardroom I test new models as they land, with Hermes Agent courses and four coaching calls a week. → Get access here
How To Set Up Solar Mini 4 In Hermes Agent
Setup is quick, and there are two routes.
Hermes Cloud
- Go to portal.nousresearch.com/cloud and create an agent.
- Open the model section and scroll down.
- Click "Free" and choose Upstage Solar Mini 4.
Hermes Desktop
- Create a new profile and name it "solar mini 4".
- Choose Nous Portal, refresh the models list and pick Solar Mini 4 from the free models.
- Open a new chat and send a test message.
I keep one profile per model, which makes fair comparisons easy.
For every click and screen, use my Solar Mini 4 Hermes setup guide.
The free-vs-paid trap
This is the one thing you must do right.
Run hermes dashboard, or open it from the Agent OS manage section, then go to Models and click "Set main model".
Type "solar", choose Nous Portal as the provider, and pick the free Solar Mini 4 rather than the paid one.
The free ID ends in ":free", and the paid one quietly bills you.
What Else Is On The Free List Next To Solar Mini 4?
When I filtered Nous Portal by "Free", Solar Mini 4 sat beside NVIDIA Nemotron, Poolside Laguna XS 2.1, inclusionAI Ling 3.1 Flash and Ling 3.0 Flash, StepFun Step 3.7 Flash and Meituan LongCat 2.0.
That list changes, so check it on the day.
If Solar Mini 4 gets rate limited, any of those makes a sensible fallback profile.
For how bigger coding models stack up, see my ranking of the best AI models for coding.
Strengths That Earned Solar Mini 4 Its Score
- It's genuinely fast, which makes chatting with an agent feel snappy.
- It's free on Nous Portal for two weeks, so testing costs you nothing.
- It handled a real research task with sources and a useful summary.
- It reads very long inputs thanks to the 512K context window.
- It supports tool calling and structured outputs, which agents need.
- It works in Korean, English and Japanese.
Weaknesses That Cost Solar Mini 4 Points
- It isn't frontier level, and it's well behind Claude Opus 5.5 on hard work.
- It's very verbose according to Artificial Analysis, which matters once you pay per token.
- Its Terminal-Bench 4.0 score of 1% says it's a poor pick for terminal coding agents.
- Free models get rate limited, and the Nous Portal window only lasts two weeks.
- The weights aren't public, so you can't run it offline.
- One viewer said it ran badly for them, so results vary with the task.
Who Should Use Solar Mini 4?
It suits Hermes Agent users who want a free, fast second profile for news checks, summaries and drafts.
It suits businesses working in Korean, Japanese or English that want cheap long-document reading.
It doesn't suit anyone who needs top-tier coding, deep strategy or terminal automation.
If you're weighing up harnesses as well as models, my DeepSeek Harness review covers that side.
🔥 Want the full Hermes setup with every model profile ready to go? The AI Profit Boardroom has the step-by-step Hermes training and a community of 3,400+ members comparing models every week. → Join the AI Profit Boardroom
Related Reading
- The Solar Mini 4 Hermes setup shows every step in Hermes Cloud, Hermes Desktop and the dashboard.
- The best AI models for coding ranks the models I'd use for serious builds.
- My DeepSeek Harness review scores another free-friendly agent setup.
Also On Our Network
- 🌐 Step-by-step: running Solar Mini 4 free as an operator
- 🌐 Seven business uses for Solar Mini 4
- 🌐 The Solar Mini 4 quick start on agentos.guide
- 🌐 Solar Mini 4 against real GoldieBench scores
Solar Mini 4 Verdict FAQ
Is Solar Mini 4 any good?
It's good for fast, basic agentic tasks like research summaries and drafts, and I score it 3.5 out of 5.
It isn't frontier level, so keep a stronger model like Claude Opus 5.5 for hard work.
What are Solar Mini 4's specs?
Upstage lists 35 billion total and 3 billion active parameters, a 512K context window, up to 128K output tokens, and Korean, English and Japanese support.
It supports reasoning mode, tool calling and structured outputs, and it scores 24 on the Artificial Analysis Intelligence Index.
Is Solar Mini 4 free?
It's free on Nous Portal for a two-week window inside Hermes Agent, and the Upstage Console Playground lets you try it with no API key.
Otherwise it costs $0.10 in and $0.40 out per million tokens on Upstage, or $0.05 and $0.20 on OpenRouter.
Is Solar Mini 4 open source?
No, it isn't open source.
It's a proprietary model and Upstage hasn't released the weights, so you can't run it locally.
What is Solar Mini 4 bad at?
Artificial Analysis recorded 1% on Terminal-Bench 4.0 and flagged it as very verbose.
It also struggles with hard reasoning compared with frontier models, and free access can be rate limited.
Who makes Solar Mini 4?
Upstage makes Solar Mini 4.
Upstage is an AI company from South Korea, and Solar Mini 4 is the compact sibling of its Solar Pro 4 model.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure agency, Goldie Agency, with a team of around 50 people.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote the Amazon best-sellers "SEO Link Building Mastery" and "Agency Marketing Mastery".
- My Udemy courses have taught over 50,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Final Verdict On Solar Mini 4
My final score is 3.5 out of 5, and the free window makes it an easy one to try.
Use it for speed and cheap everyday jobs, keep a stronger model for the hard work, and you'll get real value out of Solar Mini 4.
