GPT-6 Luna vs Hy4
Pricing, context window and real answers (November 2026)
GPT-6 Luna (OpenAI) and Hy4 (Tencent), compared as you can actually call them on FastMetal. Both are served from the same OpenAI-compatible endpoint and API key; switching is a change to the model string.
Specs and pricing
 GPT-6 Luna |  Hy4 | |
|---|---|---|
| Provider | OpenAI | Tencent |
| Input (per 1M tokens) | $0.1055 | $0.8799 |
| Output (per 1M tokens) | $0.5275 | $2.64 |
| Typical cost (1,000 in + 500 out tokens × 1,000 calls) | $0.3693 | $2.20 |
| Context window | 1,050,000 tokens | 1,000,000 tokens |
| Release date | 9/22/2026 | 11/4/2026 |
| Input modalities | file, image, text | text |
| Throughput (p50) | 97 tokens/s (n=24) | — |
| Arena · overall | #97 · ELO 1,441 | Unranked |
| Arena · Japanese | #73 · ELO 1,421 | Unranked |
| Arena · coding | #66 · ELO 1,506 | Unranked |
Prices are FastMetal's rates, before tax.
Rankings: LMArena leaderboard dataset (lmarena-ai/leaderboard-dataset), CC BY 4.0, snapshot 2026-10-28; category names mapped and ratings rounded by FastMetal.
Which should you pick?
- On price, GPT-6 Luna: $0.5275 per 1M output tokens, 80% less than Hy4.
- For long documents, GPT-6 Luna: a 1,050,000-token context window against 1,000,000 for Hy4.
- If in doubt, try both on the same key. Switching is a change to the model string, and each is billed at its own rate.
Real answers to the same prompts
Exactly what the FastMetal gateway returned, side by side. Not benchmark scores: actual output.
Count the number of 'r's in 'strawberry'
Count the number of 'r's in 'strawberry'. Explain your reasoning step by step.
GPT-6 Luna
In **“strawberry”**, the letter **r** appears at positions 3, 8, and 9. So there are **3** “r”s.
2.7s94 output tokens
Hy4
The word **“strawberry”** is spelled:
**s t r a w b e r r y**
Step by step:
1. Check each letter: s, t, **r**, a, w, b, e, **r**, **r**, y
2. Mark the **r**’s: one after **t**, and two near the end before **y**.
3. Count them: 1 + 1 + 1 = **3**
So, there are **3** ‘r’s in **“strawberry”**.
4.2s343 output tokens
Debug This Error
I'm getting the following error in my Node.js application: TypeError: Cannot read properties of undefined (reading 'map') at UserList (/app/components/UserList.js:12:25) at renderWithHooks (/app/node_modules/rea…
GPT-6 Luna
`users` is `undefined` when `UserList` renders, so calling `users.map(...)` throws. This often happens because the parent didn’t pass the prop or because data hasn’t loaded yet.
If an empty list is acceptable until the data arrives, provide a default:
```jsx
function UserList({ users = [] }) {
return (
<ul>
{users.map(user => (
<li key={user.id}>{user.name}</li>
))}
</ul>
);
}
```
Also check that the parent passes `users`:
```jsx
<UserList users={users} />
```
If `users` is loaded asynchronously, initialize it as an array:
```jsx
const [users, setUsers] = useState([]);
```
If `null` is also possible, normalize it before mapping, for example `(users ?? […
3.8s289 output tokens
Hy4
The error is caused by this line:
```js
{users.map(user => <li key={user.id}>{user.name}</li>)}
```
At render time, `users` is `undefined`. Since `.map()` only exists on arrays, JavaScript tries to read `map` from `undefined` and throws:
```txt
Cannot read properties of undefined (reading 'map')
```
Even though this is in a Node.js app, this specific error is happening during React rendering, likely server-side rendering or a Node-based build/render process.
Common causes:
1. The parent component is not passing `users`.
2. The prop name is misspelled.
3. `users` comes from async data and is `undefined` on the first render.
4. Your API returns something like `{ users: [...] }`, but you…
20.3s1951 output tokens
Code Review
Please review the following Python function and suggest improvements for readability, performance, and best practices: def get_data(url, retries=3): import requests import time for i in range(retries):…
GPT-6 Luna
Key improvements:
- Move imports to module scope.
- Catch `requests.RequestException` instead of using a bare `except`; the current handler can hide programming errors as well as network failures.
- Call `raise_for_status()` so unsuccessful HTTP responses are handled explicitly.
- Use a list comprehension instead of building the result manually.
- Replace `item["active"] == True` with `item.get("active") is True` if the field is expected to be the Boolean `True`.
- Add backoff between attempts. A fixed two-second delay can make retries slower than necessary.
- Decide whether failures should raise an exception or return `[]`. Returning `[]` makes a failed request indistinguishable from a suc…
16.5s1713 output tokens
Hy4
Here are the main improvement areas, followed by a possible rewrite.
## Key issues
### Readability / style
- Move imports to the top of the module.
- Use descriptive names: `response` instead of `r`, `attempt` instead of `i`.
- `if item["active"] == True:` should be `if item.get("active") is True:` or simply `if item.get("active"):`, depending on intent.
- The filtering loop can be a list comprehension.
- Add type hints and a docstring.
### Performance
- Repeated imports are cached by Python, so not a major cost, but top-level imports are cleaner.
- Avoid sleeping after the final failed attempt.
- Use `requests.Session` for connection pooling if this function may be called repeatedly or i…
110.2s10761 output tokens
Frequently asked questions
- Which is cheaper, GPT-6 Luna or Hy4?
- Per 1M output tokens, GPT-6 Luna is $0.5275 and Hy4 is $2.64 on FastMetal (US dollars, before tax), so GPT-6 Luna is cheaper.
- How do the context windows of GPT-6 Luna and Hy4 compare?
- GPT-6 Luna takes 1,050,000 tokens; Hy4 takes 1,000,000.
- Can I use GPT-6 Luna and Hy4 with the same API key?
- Yes. On FastMetal's OpenAI-compatible endpoint, switch by passing "gpt-6-luna" or "hy4" as the model. Each is billed at its own rate from the same prepaid balance.
Try both on one API key
Create an account and add credit to call GPT-6 Luna and Hy4 from the browser chat and the API. No monthly fee.