I ────── The Tencent + Hermes Machine™

Free until ~21 July 2026 · dropped 7 July

The Tencent + Hermes Machine™.

Tencent just dropped Hy3 — a 295-billion-parameter model built for AI agents. It's free for two weeks. I wired it into Hermes and tested it. Here's everything.

Hermes, fully clothed in a flowing golden chiton and winged helmet, standing on the head of a great jade-and-gold Chinese dragon machine, holding its reins, with a glowing hourglass beside them

Hy3 is Tencent Hunyuan's newest brain: 295B parameters, built for cost-effective agentic work — coding, reliable tool-calling, reasoning, and a 256K context window. Nous Research put it on their Portal free for the next two weeks. That means the same Hermes agent that runs my Agent OS can drive a 295B frontier-class model right now, for exactly $0. I set it up, I tested it, and every receipt is below — plus the exact commands so you can do it in five minutes.

295Bparameters (MoE)

$0for two weeks on Nous Portal

256Kcontext window

2/2my real tests passed first try

II ────── How it works, in simple words

A giant brain, a free doorway, and your agent.

Three pieces. That's the whole system.

1. The brain: Hy3. Tencent's new model. It's a "Mixture of Experts" — 295 billion parameters total, but it only wakes up the experts it needs for each question. That's why it's fast and cheap to run despite the size. It was tuned for one job: being a reliable worker inside AI agents.

2. The doorway: Nous Portal. One API that serves 268 models — Claude, GPT, Gemini, and now Hy3. You log in once, and any tool that speaks to the Portal can use any model on it. For two weeks, Hy3's meter reads $0.

3. The agent: Hermes. Hermes is Nous's own agent CLI — the one already inside my Agent OS. Point a Hermes profile at Hy3, and every task you hand Hermes gets done by a 295B brain instead of whatever you were paying for before.

The whole machine — Hermes hands out the work, the Portal opens the door for free, Hy3 does the thinking.

III ────── Exactly how it works, step by step

What happens when you run it — no mystery.

You log Hermes into Nous Portal once. hermes auth add nous runs a device login — a browser opens, you approve, and Hermes stores the token. If you've logged in before, it finds the saved credentials and reuses them. No API key to copy, ever.

You create a profile pointed at Hy3. A Hermes profile is an isolated copy of the agent with its own model and settings. Mine is called hy3, and its config says: model tencent/hy3:free, provider nous. The :free suffix matters — that's the two-week free tier. The plain slug is the paid meter.

You type a task. hy3 -z "build me a galaxy page". The wrapper alias picks the profile, the profile picks the model.

Hermes bundles your task with its tools. The prompt goes out with the tool list — file tools, shell tools, search — so Hy3 can act, not just talk. That's the agentic part.

The Portal routes it to Hy3 for $0. Your request hits Nous Portal, the Portal checks the free tier, and forwards it to Tencent's 295B model. Only the needed experts inside the model wake up — that's the MoE trick that keeps it fast.

Hy3 answers — or acts. Plain questions come back as text. Agent tasks come back as tool calls: Hermes executes them (write the file, run the command), sends the results back, and Hy3 keeps going until the job is done.

You keep every other model. Nothing about your setup changes. Hy3 is one more profile beside your others — switch by typing a different alias. When the two weeks end, keep it (paid), or delete the profile in one command.

One task's journey — typed in your terminal, packed by Hermes, routed free by the Portal, executed by Hy3.

So when someone asks "but what IS this?" — the honest one-line answer: it's one login and one small config file that point your existing Hermes agent at Tencent's newest 295B brain, free for two weeks.

The MoE trick in one picture — most of the brain sleeps, so the giant runs at small-model speed.

IV ────── Straight from the source

The announcement. Read it and run it yourself.

The announcement · 7 July 2026

Nous drops the news

This is where it started. Nous Research announcing Hy3 — Tencent Hunyuan's new 295B Mixture-of-Experts model — free in Nous Portal for two weeks. Read the framing: "cost-effective agentic use… coding, tool-calling reliability, reasoning, and 256K long-context tasks." That's not a chatbot pitch. That's a worker-model pitch — exactly what you plug into an agent like Hermes.

The second door · same day

It's free in Kilo Code too

Hours before the Nous drop, Kilo Code made HY3 free as well — "for a limited time" inside their coding agent. They ran the fun test themselves: one-shotting a 3D spinning-Earth visualizer with Opus, Fable and HY3 side by side in Kilo's Agent Manager. So there are two free doors into the same 295B brain right now: Hermes via Nous Portal (this guide's setup), or Kilo Code if that's your editor. Same dragon, pick your door.

Official sources + everything I built ↓

the announcementNous Research on X → also free hereKilo Code's HY3 alert → the doorway · sign upNous Portal → the agentHermes Agent docs → the model makerTencent Hunyuan on X → my receipt · live buildThe galaxy Hy3 built me →

V ────── My story · why this matters

I was you. Then I bolted a 295B brain into my agent for free.

Before

Every frontier model I wanted to test meant another API key, another meter running.

Free models were either tiny, throttled, or fell over on tool calls.

My agents needed a reliable worker model — and reliable always meant paid.

So new releases sat on my "try it someday" list while the trial windows closed.

Then Hy3 landed on Nous Portal — free — and Hermes already had the doorway built in.

After

Five minutes of setup. One login, one profile.

Now a 295-billion-parameter model does my agent work — and the meter reads $0.

It one-shotted a build test and passed a tool-calling test on the first try.

And when the two weeks end, I'll know exactly whether it earns a paid slot — from receipts, not reviews.

You can have this too. Same tools. Five minutes.

VI ────── The receipts

Real people. Real wins. Inside the Boardroom right now.

3,600+ Founders inside AIPB
400k YouTube subscribers
38 Countries · live members
163k X / Twitter followers
29k+ Udemy students

I'm not going to paste invented quotes here. The wins are real and written by the members themselves — agency owners, ecom founders, course creators, solo operators across 38 countries. Read them in their own words.

Read the 158-page wins doc →

Before you scroll on —

Commit to transitioning today. Not tomorrow.

This one has a clock on it. Two weeks, then the meter starts.

The next five minutes show the exact setup — every command, tested this morning.

So here's the deal.

If you're reading this — promise yourself one thing right now. You're going to finish this guide AND run the login command before you sleep tonight. Just one login. Because the moment a frontier-class model starts doing your agent work for free, you stop guessing what these models can do and start knowing.

The people sitting still will read the reviews after the window closes. The people implementing today get two free weeks of receipts.

Be one of those people.

Commit to the transition. Commit to taking action today. This changes everything about your workflow.

VII ────── I tested it — here's the proof

My real tests: both passed first try.

Tencent's pitch is agentic reliability — coding and tool-calling. So I tested exactly those two things through my new Hermes profile.

Test 1 — one-shot coding. I asked for a complete single-file animated galaxy page — 2,000 stars, canvas 2D, no libraries, mouse parallax. One shot, 7,290 bytes, rendered perfectly, zero console errors:

A colorful spiral galaxy with a glowing pink core and the title HY3 GALAXY, built by Hy3 in one shot

Hy3's one-shot build — headless render check passed, no errors. Open the live version from the sources panel above.

Test 2 — tool-calling. I told it to use its file tools: create a file with exact content, read it back, report. It called the write tool, called the read tool, and reported back: "reading it back returns exactly 'tool call works'" — even noting the byte count. Precise, no fumbling, no fake tool syntax. This is the thing cheap models fail at, and it's the thing Hy3 was built for.

"Free for two weeks — what's the catch?"

The catch is the calendar. It's a launch promo: Tencent and Nous want builders to try the model before the meter turns on around 21 July. Your data terms are Nous Portal's standard terms — same as using any model there. The honest play: use the window hard, keep the receipts, and decide with evidence whether it earns a paid slot after.

VIII ────── The exact setup

Wire it into Hermes — 5 minutes, 3 commands.

This is everything I ran, cleaned up so it works first time.

# 1. create an isolated profile for Hy3
hermes profile create hy3 --description "Tencent Hy3 295B via Nous Portal"

# 2. log the profile into Nous Portal (browser opens once — or reuses saved creds)
hy3 auth add nous

Then set the model. Open ~/.hermes/profiles/hy3/config.yaml and make the model block read:

model:
  default: tencent/hy3:free
  provider: nous
  context_length: 256000

And test it:

hy3 -z "Reply with exactly: HY3 READY"
# → HY3 READY

Two gotchas I hit so you don't have to:

Gotcha 1 — the :free suffix. The plain slug tencent/hy3 is the PAID tier — you'll get "requires available credits". The free two-week tier is tencent/hy3:free. One suffix, whole difference.

Gotcha 2 — create the profile properly. Use hermes profile create, don't hand-make the folder. The create command scaffolds the profile and gives you the hy3 wrapper alias; a hand-made folder leaves Hermes saying "No LLM provider configured".

"I'm not technical. Is this really 5 minutes?"

Three commands and one small text file. The login opens a normal browser page — you click approve, that's it. And if you're inside the Boardroom, the Agent OS zip ships with profiles pre-configured — you skip even these steps.

"Am I sending my data to Tencent?"

Your requests go to Nous Portal, and the Portal serves the model — you sign up under Nous's standard terms, same as using any of the 268 models there. The honest rule I use for every provider, free or paid: launch promos are for testing capability, so I throw builds and public work at it — not client secrets. Decide your own line before you start, then use the window hard inside it.

256K of free context — a quarter-million tokens per prompt, 8× a typical free model.

IX ────── The framework

The Tencent + Hermes Machine™.

Three parts, one machine. Each part removes a cost or a limit.

The Brain

You stop paying for frontier-class thinking — Hy3's 295B Mixture-of-Experts does the reasoning, coding and tool-calling, and for two weeks its meter reads zero.

ii.

The Doorway

You stop juggling API keys — Nous Portal is one login for 268 models. Today's free giant is Hy3; when the next one lands, the same doorway opens to it.

iii.

The Hands

You stop copy-pasting model output — Hermes turns Hy3's answers into actions: files written, commands run, builds shipped, straight from your Agent OS.

"Doesn't running Agent OS burn a fortune in tokens?"

No — that's the biggest myth about it, and this guide is the proof. Agent OS runs the everyday 90% on free local models, plugs in free API windows exactly like this one — a 295B model at $0 — and drives the CLIs you already pay for only on the hardest work. Add the Caveman token-diet on top and even the paid replies shrink 60–75%. Inside the AI Profit Boardroom there are full token-optimisation tutorials, so you cut usage to the bone and never think about it again.

X ────── Old way vs new way

Old way vs new way.

Old way ~1 hr of signup, still paying

See a model launch on X, add it to the "someday" list
New account, new API key, new billing page
Copy the key into three tools, break one of them
Pay per token from minute one just to evaluate it
Free alternatives throttle or botch tool calls
Trial window closes before you ever ran a real task

New way ~5 min, $0 for two weeks

One hermes auth add nous login, reused forever
One profile file points at tencent/hy3:free
A 295B frontier-class model doing real agent work, free
Tool calls that actually work — tested, first try
Same doorway serves the NEXT free model with one slug swap
Decide with two weeks of receipts, not reviews

"Free models always throttle me — won't this one?"

Fair scar — most free tiers do. In my tests today, Hy3 answered promptly and never rate-limited me: smoke test, a full one-shot build, and a tool-calling round-trip, back to back. Launch windows are usually the most generous a model will ever be — the whole point is showing off capacity. If limits appear as the window fills up, the receipts you collect early still tell you whether it's worth paying for.

Two free weeks, one decision at the end — the maths never gets better than day one.

Get the whole operating system

Want this pre-wired, plus the whole machine room?

The Tencent + Hermes Machine is one profile inside the Agent Operating System — the dashboard I run my whole business on. Join the AI Profit Boardroom and you get everything:

Hermes pre-configured — profiles, Portal login flow, and every free-model window as it opens

Free local models — the everyday 90% of work at $0 on your own machine

The Caveman Command Engine — 60–75% off every paid reply

Every CLI you already pay for — Claude, Codex, Gemini, Kimi, GLM, Grok in one dashboard

Hermes Astros — the 24/7 YouTube competitor watcher that writes your titles

The Hermes Oracle — its sibling that watches X and drafts your posts

The Video Director — topic in, finished video out

Agent Kanban — Planner → Builder → Reviewer teams that ship real work

The memory vault — an Obsidian brain your agents actually read

3,600+ founders + me — daily tutorials, weekly calls, new tools added the week they ship

You're not buying a tool. You're getting the whole operating system I run a seven-figure business on — as a zip, with coaching calls where we set it up together.

Get the Agent OS →

Inside the AI Profit Boardroom · skool.com/ai-profit-lab

Set up in an afternoon · used in 38 countries · new tools added every week

XI ────── Three beliefs to drop

What's holding you back.

Wrong: "Free models are toys — the real work needs the expensive ones."

Right: Hy3 is a 295B frontier-class model that one-shotted my build test and nailed tool-calling. "Free" here isn't a smaller brain — it's a launch window on a big one. The skill is catching the windows.

Wrong: "I'll try it when I have time — these offers come around again."

Right: This one has a date on it: two weeks from 7 July. After that the meter turns on. Five minutes of setup today buys fourteen days of a free 295B worker — the maths never gets better than that.

Wrong: "Another model means another stack of keys and billing pages."

Right: That's exactly what the Portal kills. One login, 268 models, and Hermes already speaks to it natively. Adding Hy3 was a two-line config file — and swapping to the next model is one slug.

Don't take my word for it

158 pages of members who already broke through these exact beliefs. Their stories — real businesses, real wins — are documented here.

Read the 158-page testimonials doc →

XII ────── The SOP

Your two free weeks — an 8-step plan.

Sign up at portal.nousresearch.com. Free account — this is the doorway everything else uses.

Create the profile. hermes profile create hy3 then hy3 auth add nous — approve the browser prompt once.

Set the model. tencent/hy3:free, provider nous in the profile's config.yaml. Remember: the :free suffix or you'll hit the paid meter.

Smoke test. hy3 -z "Reply with exactly: HY3 READY". Two seconds, confirms the whole chain.

Throw your real work at it. Builds, research, refactors — the tasks you'd normally give your paid model. Its strengths: coding, tool-calling, long context.

Use the 256K window. Feed it whole codebases and long documents — that's a quarter-million tokens of context, free.

Keep score. Note what it nails and what it fumbles versus your current model. Two weeks of receipts beats any benchmark chart.

Decide on day 13. If it earned a paid slot, drop the :free suffix. If not, delete the profile — hermes profile delete hy3 — and you spent nothing either way.

XIII ────── Recap

What you gain.

You got a 295B brain for $0. Frontier-class reasoning, coding and tool-calling, free for two weeks on Nous Portal.

You skipped the key-juggling. One device login, no API keys copied anywhere, credentials reused automatically.

You kept your whole setup. Hy3 is one more profile beside your others — nothing replaced, nothing broken.

You saw real receipts. One-shot galaxy build, perfect tool-call round-trip — both passed first try, both shown above.

You dodged the gotchas. The :free suffix and the profile-create trap — solved here so your setup works first time.

You got 256K of context. Whole codebases and long documents in one prompt, at zero cost.

You own the doorway. The same Portal login serves the next free window — and if Kilo Code is your editor, HY3 is free there too right now. Two doors, same dragon.

You'll decide with evidence. Two weeks of real work, then keep it or delete it — receipts, not reviews.

Your move

The window is open right now.

Every few weeks a door like this opens — a frontier model, free, for days. The people with the Agent OS walk through the same afternoon, because the machine room is already built: Hermes wired to the Portal, free local models for the everyday work, the Caveman engine shrinking every paid reply, Astros watching YouTube, the Oracle watching X, the Video Director and the Kanban teams shipping while you sleep. That's what you get as a zip file — with coaching calls where we set it up together, step by step. Daily tutorials. A 30-day roadmap. 3,600+ founders across 38 countries, someone online whenever you get stuck. And when the next window opens? I test it, I break it, and it's in the OS the same week.

Join the AI Profit Boardroom →

Inside the AI Profit Boardroom · skool.com/ai-profit-lab

Keep every file if you leave · 7-day refund · the only way you lose is by not trying

I'll see you in the next one.