⭐ Get the Agent OS + join 3,400+ founders inside the AI Profit Boardroom → Join AIPB ($69/mo)

GoldieBench Blog · 9 min read

Hermes MissingNo: What The Free Stealth Model Is, How To Run It, And How It Compares With The Benched Free Models

Hermes MissingNo explained: what the free stealth model is, how to enable it, my test results, and how the benched free Nous models really score.

Hermes MissingNo: What The Free Stealth Model Is, How To Run It, And How It Compares With The Benched Free Models — illustrated hero

Hermes MissingNo is a new free stealth model that you can run as the brain of Hermes Agent through Nous Portal.

MissingNo is a coding and agentic reasoning model, and Nous Research announced it on X on 10 October 2026.

Nous Research make Hermes Agent, and they have made MissingNo free for a limited time.

The lab that built the model hasn't been named, which is what makes it a stealth release.

In this post I'll explain what it is, what it costs, how to enable it, how to use it as a Hermes profile and how it did in my own tests.

My MissingNo video isn't public yet, so the video above is my Solar Mini 4 walkthrough, which shows the same Hermes Cloud free-model screens.

What Hermes MissingNo is

Hermes MissingNo means setting MissingNo as the main model on a Hermes Agent profile.

Hermes Agent handles the tools, memory, skills and chat, and the model does the thinking.

The name is a Pokémon reference, because MissingNo is short for "missing number", the glitch Pokémon from the original games.

Nous Research's announcement opened with "Wild MISSINGNO. appeared!", so the joke is official.

Here's what I could verify from that post and from the Nous inference API on 12 October 2026.

FactDetail
Released byNous Research made it available on Nous Portal and announced it on 10 October 2026.
Built byThe lab hasn't been announced.
Described asIt's described as a coding and agentic reasoning model that is highly token efficient.
CostIt's free for a limited time, with no published end date.
Model IDThe Nous inference API lists it as stealth/missingno.
Context windowThe listing shows 262,144 tokens, with up to 131,072 output tokens.
InputsIt takes text only.
ToolsThe listing shows tool calling, parallel tool calls and structured outputs.
ReasoningReasoning is always on, with low, medium, high, xhigh and max levels.
Not publishedThe maker, the parameter count and official benchmarks aren't published.

I've guessed that it could be Nous Research's own model, and I want to be clear that it's only a guess.

The company background is in my Hermes Agent by Nous Research post.

What MissingNo costs

MissingNo costs nothing today.

The Nous inference API showed a price of zero for input and output when I checked.

Nous Research said it's free for a limited time, and they didn't give an end date.

A stealth model can be removed or turned into a paid model at short notice.

So I use it as a free extra, and I always keep a second model ready.

How to enable MissingNo in Hermes Cloud

Hermes Cloud runs your agent on Nous Research's servers, so there's nothing to install.

StepWhat to do
1Go to portal.nousresearch.com/cloud and click Deploy.
2Log in to Nous Portal.
3Click Create agent.
4Open Manage agents.
5Click on your new agent.
6Click Open dashboard.
7Go to the model section.
8Search "Nous Research", or type "missing".
9Pick the stealth model, which is MissingNo.
10Reload the page and start chatting.

You can also ask Hermes to configure Hermes Cloud for you, and it will change the model itself.

I'd still check the model section afterwards, so you know the change went through.

When I checked the API list I saw one MissingNo entry priced at zero, so I didn't see a paid twin next to it.

How to use MissingNo as a Hermes profile

I keep one Hermes profile per model inside my Agent OS.

I have a Step 5 Preview profile, which is another free model, and a MissingNo profile beside it.

Each profile has its own model, settings and memory, so testing one never breaks another.

On your own machine, hermes profile create missingno makes the profile.

The hermes dashboard command opens the dashboard where you set the main model, and hermes -p missingno opens a chat in that profile.

If MissingNo is missing from the picker, hermes model --refresh re-fetches every provider's live model list.

How MissingNo did in my Hermes tests

I ran it through the kind of jobs I actually give an agent.

TestWhat happenedMy take
/learnIt created a skill really well and fast, and it named the path.This was the best result.
CodingA simple ping-pong style game took about five minutes, and it works.It's moderate, and it's not frontier level.
Tool callingIt was fast at calling agentic tools.It suits agent loops.
Reply speedReplies were quick on Hermes Cloud.Hosted models respond fast.

The /learn command points Hermes at a source and turns it into a reusable skill.

MissingNo did that really well and fast.

The coding task didn't work so well, because a simple game took around five minutes to generate.

The game works, but it won't replace Claude Opus 5.5.

Those are my own hands-on results, and they are not GoldieBench scores.

Where else MissingNo works in Hermes

The model follows the profile, so it works across Hermes.

You can use it in Hermes Bots by creating a bot from your active profile.

You can use it in Hermes Oracle, where I'd start with competitor analysis.

You can use it in other workflows that read the model from the profile.

I also demoed a pop-up voice agent in my Agent OS that works with ChatGPT and Hermes.

I asked it for useful Hermes commands, and it listed /learn and /skills.

The bot feature has its own explainer in Hermes Agent Bot Mode.

What to do if MissingNo is removed or becomes paid

MissingNo may be taken down, it may become paid, or you may run out of tokens.

When that happens, open the model section and type "free".

There were about eight free models when I recorded, and these are the ones I named.

Free modelMakerWhen I'd switch to it
Step 3.7 FlashStepFunI'd use it as a quick everyday backup.
Step 5 PreviewStepFunI already keep a profile for it.
Ling 3.0inclusionAII'd use it as a second free flash option.
LongCatMeituanI'd use it when the input is very long.
Laguna XS 2.1PoolsideI'd use it for light coding chores only.
Solar Mini 4UpstageI'd use it for fast research, on the free entry.
MissingNoUnannouncedIt's the stealth model this post is about.

The free list changes often, so check it inside Hermes on the day.

The limits to know before you rely on it

MissingNo takes text only, so image jobs need a different model.

Reasoning is always on, so short questions can take a moment longer.

Nobody has published a parameter count or an official benchmark for it.

The free window has no published end date, and it can close without warning.

Like any free model, it can also run out of tokens partway through a job.

Coding was only moderate in my test, so don't make it your main builder.

Is MissingNo on GoldieBench?

MissingNo is not on the GoldieBench leaderboard, so I'm not going to give it a score it hasn't earned.

GoldieBench is my leaderboard where every model gets the same one-shot prompts to build games, pages, simulations and visuals.

Each build is rendered and scored from 0 to 10, and the number you see is the average across those tasks.

That measures how well a model plans and writes working code in one go.

It doesn't directly measure agent work like /learn or tool calling inside Hermes.

So a model can be handy in an agent and still be a weak one-shot builder, and my MissingNo tests point that way.

One slow game build is a single hands-on test, and it isn't a measured score.

I'll only put a number next to MissingNo if it goes through the same tasks as every other model.

How I'd test MissingNo against the board yourself

You don't have to wait for me to bench it.

You can run the same style of test in your own Hermes profile in a few minutes.

First, open a fresh chat in your MissingNo profile, so no old context helps it.

Second, ask for one self-contained HTML page, such as a small game or a calculator, in a single prompt.

Third, time how long the answer takes, because my game took about five minutes.

Fourth, open the page and check that it loads and the main control works first time.

Fifth, give the same prompt to a second free model and compare the two results side by side.

That's the spirit of a one-shot build test, because the model gets one go and no hand-holding.

If the page works first time, the model has passed a test that many small models fail.

If it needs three rounds of fixes, keep it on agent chores and away from builds.

The free Nous models that are benched

Two of the free models on that list do have real GoldieBench numbers.

Free modelOn GoldieBench?Live avgScored tasksWhat that tells you
MissingNoNoNot scored0It isn't benched, so my hands-on test above is all I can share.
LongCat-2.0Yes, provisional8.124It scored high on a small sample, so it's promising and not yet proven.
Laguna XS 2.1Yes3.9342It struggles on one-shot builds, so keep it on light chores.
Step 5 PreviewNoNot scored0It isn't benched, so test it on your own jobs.
Step 3.7 FlashNoNot scored0It isn't benched, so test it on your own jobs.
Ling 3.0NoNot scored0It isn't benched, so test it on your own jobs.
Solar Mini 4NoNot scored0It isn't benched, so test it on your own jobs.

LongCat-2.0 averages 8.12, but only 4 tasks are scored, which is why the board marks it provisional.

Laguna XS 2.1 averages 3.93 across 42 tasks, which is a much bigger sample and a much weaker result.

Those two numbers show how wide the gap between free models can be.

Being free tells you nothing about quality, so you have to test.

Where MissingNo would likely sit

I haven't benched MissingNo, so I won't guess a number.

I can tell you how I'd think about it.

My coding test produced a working game, so it can ship something that runs.

It also took five minutes and wasn't frontier level, so I don't expect it to trouble the top of the board.

For context, Claude Opus 5.5 averages 7.57 across 50 tasks, and I said plainly that MissingNo won't replace it.

Claude Opus 5 averages 8.27 across 50 tasks, and MiniMax M3 averages 7.97 across 47 tasks.

Those are the kinds of models I'd give real builds to.

MissingNo's strengths in my tests were skills and tool calls, and GoldieBench doesn't score those.

The Hermes brain plan I'd use today

LaneBrainWhy
Skills and tool-heavy agent jobsMissingNo while it's freeIt was great at /learn and fast at tool calls in my test.
Free and heavierLongCat-2.0 (8.12, provisional)It scored high on a small sample, so it's worth testing on bigger jobs.
Light free choresLaguna XS 2.1 (3.93)It's free, but it's weak on one-shot builds.
Value buildsMiniMax M3 (7.97) or GLM-5.2 (7.77)They get near-frontier scores at low prices.
Strongest buildsClaude Opus 5 (8.27)It has the best single-model average on the board.
LocalQwable 5 27B Coder (7.14)It's the best local builder on the local board.

Hermes doesn't ship with one fixed model, so the brain you set on each profile decides how well it works.

The smart move is to give each profile the cheapest brain that can actually do its job.

My wider picks are in the best Hermes Agent model and the best LLMs for Hermes Agent.

If you want another free model walkthrough, Solar Mini 4 Hermes covers one from the same list.

Also On Our Network

Use it for skills and tool calls while it's free, keep a benched backup ready, and watch the board for the day a real score lands for Hermes MissingNo.

FAQ

What is MissingNo in Hermes Agent?

MissingNo is a stealth coding and agentic reasoning model that Nous Research announced on 10 October 2026. It is free for a limited time on Nous Portal, and the lab that built it has not been named.

Is MissingNo on the GoldieBench leaderboard?

No. MissingNo has not been benched on GoldieBench, so it has no score. Julian's hands-on test showed strong /learn results, fast tool calling and a moderate coding result, but that is not a benchmark score.

How do I enable MissingNo in Hermes Cloud?

Click Deploy at portal.nousresearch.com/cloud, log in, create an agent, open Manage agents, click the agent, open its dashboard, go to the model section, type missing, pick the stealth model, then reload and chat.

Which free Nous models are benched on GoldieBench?

LongCat-2.0 is benched provisionally at 8.12 across 4 tasks, and Laguna XS 2.1 averages 3.93 across 42 tasks. MissingNo, Step 5 Preview, Step 3.7 Flash, Ling 3.0 and Solar Mini 4 are not benched.

Does Hermes Agent come with its own model?

No. Hermes Agent runs whichever model you set as a profile's main model. You can pick a free model such as MissingNo on Nous Portal, a paid model or a local model, and swap by changing that one setting.

Is MissingNo good for coding?

It is moderate. In Julian's test a simple ping-pong style game took about five minutes to generate and it worked, but it was not frontier level and it will not replace Claude Opus 5.5.

The same stack Julian uses

Run this stack yourself.

Every demo on this bench was built inside the Agent Operating System — one prompt, one shot, single HTML file out. The Agent OS, the prompts, the templates, the weekly walkthroughs and 3,400+ founders shipping with it every day all live inside the AI Profit Boardroom.

3,400+founders
258documented wins
38countries
$69/momonthly