GPT-6 Luna vs Inkling
Pricing, context window and real answers (September 2026)
GPT-6 Luna (OpenAI) and Inkling (Thinking Machines), 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 | Inkling | |
|---|---|---|
| Provider | OpenAI | Thinking Machines |
| Input (per 1M tokens) | ¥17.87 | ¥178.7 |
| Output (per 1M tokens) | ¥89.35 | ¥723.74 |
| Typical cost (1,000 in + 500 out tokens × 1,000 calls) | ¥63 | ¥541 |
| Context window | 1,050,000 tokens | 524,288 tokens |
| Release date | 9/22/2026 | 7/17/2026 |
| Input modalities | file, image, text | text, image, audio |
| Throughput (p50) | — | — |
| Arena · overall | Unranked | #85 · ELO 1,440 |
| Arena · Japanese | Unranked | #63 · ELO 1,415 |
| Arena · coding | Unranked | #78 · ELO 1,492 |
Prices are FastMetal's yen rates before tax. Arena ranks are as of the latest leaderboard scrape.
Which should you pick?
- On price, GPT-6 Luna: ¥89.35 per 1M output tokens, 88% less than Inkling.
- For long documents, GPT-6 Luna: a 1,050,000-token context window against 524,288 for Inkling.
- 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
Inkling
Let's break down the word **"strawberry"** letter by letter:
1. **s**
2. **t**
3. **r** ← (1st 'r')
4. **a**
5. **w**
6. **b**
7. **e**
8. **r** ← (2nd 'r')
9. **r** ← (3rd 'r')
10. **y**
So the spelling is: s-t-**r**-a-w-b-e-**r**-**r**-y
Counting them up:
- One 'r' in "str" (the 3rd letter)
- Two 'r's in "berry" (the 8th and 9th letters)
**Total: 3 'r's** in "strawberry".**
2.4s417 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
Inkling
The error occurs because `users` is `undefined` when `UserList` renders, so `users.map` fails.
This usually happens when:
- The parent doesn't pass the prop (`<UserList />` instead of `<UserList users={...} />`)
- Data is still loading (e.g., API hasn't returned yet)
## Quick Fix
Provide a default empty array so `.map` always has an array to work with:
```jsx
function UserList({ users = [] }) {
return (
<ul>
{users.map(user => <li key={user.id}>{user.name}</li>)}
</ul>
);
}
```
Alternatively, guard against undefined explicitly:
```jsx
function UserList({ users }) {
if (!users) return null; // or a loading state
return (
<ul>
{users.map(user => <li key={…
4.9s585 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
Inkling
Here is a review with an improved version followed by explanations.
### Issues in the original
- **Imports inside the function**: Re-importing on every call is inefficient.
- **Bare `except:`**: Catches `SystemExit`, `KeyboardInterrupt`, etc.
- **Silent failures**: No logging; errors disappear.
- **Magic numbers**: `10`, `2`, `200` are hard-coded.
- **Data validation missing**: Assumes `data` is a list and every item has `"active"`, `"name"`, and `"id"` (will crash with `KeyError`).
- **Performance**: Uses `.append()` in a loop instead of a list comprehension.
- **Ambiguous return**: `[]` means both "success with no results" and "total failure".
### Improved version
```python
import loggi…
54.1s5649 output tokens
Frequently asked questions
- Which is cheaper, GPT-6 Luna or Inkling?
- Per 1M output tokens, GPT-6 Luna is ¥89.35 and Inkling is ¥723.74 on FastMetal (yen, before tax), so GPT-6 Luna is cheaper.
- How do the context windows of GPT-6 Luna and Inkling compare?
- GPT-6 Luna takes 1,050,000 tokens; Inkling takes 524,288.
- Can I use GPT-6 Luna and Inkling with the same API key?
- Yes. On FastMetal's OpenAI-compatible endpoint, switch by passing "gpt-6-luna" or "inkling" 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 Inkling from the browser chat and the API. No monthly fee.