GPT-5.4 Nano vs Inkling
Pricing, context window and real answers (September 2026)
GPT-5.4 Nano (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-5.4 Nano | Inkling | |
|---|---|---|
| Provider | OpenAI | Thinking Machines |
| Input (per 1M tokens) | ¥35.74 | ¥178.7 |
| Output (per 1M tokens) | ¥223.38 | ¥723.74 |
| Typical cost (1,000 in + 500 out tokens × 1,000 calls) | ¥147 | ¥541 |
| Context window | 400,000 tokens | 524,288 tokens |
| Release date | 3/17/2026 | 7/17/2026 |
| Input modalities | file, image, text | text, image, audio |
| Throughput (p50) | — | — |
| Arena · overall | #146 · ELO 1,402 | #85 · ELO 1,440 |
| Arena · Japanese | #107 · ELO 1,368 | #63 · ELO 1,415 |
| Arena · coding | #128 · ELO 1,460 | #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-5.4 Nano: ¥223.38 per 1M output tokens, 69% less than Inkling.
- On the Overall arena board, Inkling ranks higher (#85; GPT-5.4 Nano is #146).
- On the Japanese arena board, Inkling ranks higher (#63; GPT-5.4 Nano is #107).
- On the Coding arena board, Inkling ranks higher (#78; GPT-5.4 Nano is #128).
- For long documents, Inkling: a 524,288-token context window against 400,000 for GPT-5.4 Nano.
- 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-5.4 Nano
**Step 1:** Write the word: **strawberry**
**Step 2:** Count each letter **r** in the word:
- s **t r** a w b e **r r** y
There are **3** **r**’s in total (**one** after the **t**, and **two** near the end).
✅ **Answer: 3**
1.5s86 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-5.4 Nano
This error means that **`users` is `undefined` (or otherwise not an array)** when your component renders, so calling `users.map(...)` throws:
```js
users.map(...)
^^^^ reading 'map' of undefined
```
### Why it happens
Common causes:
- You render `<UserList />` before data has loaded (e.g., `users` comes from an async fetch).
- The parent component passes the wrong prop name (e.g., `user` instead of `users`).
- The API/parent state initializes `users` as `undefined` rather than `[]`.
- You’re destructuring a prop incorrectly upstream.
### Fix options
#### 1) Ensure `users` is always an array (recommended)
Initialize/default it in the component:
```js
function UserList({ users = []…
3.3s383 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-5.4 Nano
Here are key improvements (readability, performance, robustness, best practices) and a revised version of your function.
## What to improve
1. **Avoid bare `except:`**
Catch specific exceptions from `requests` (e.g., `requests.exceptions.RequestException`) to avoid hiding bugs.
2. **Prefer `raise_for_status()` or check status codes clearly**
`r.raise_for_status()` is a common pattern.
3. **Use exponential backoff (or at least a clear retry delay)**
Instead of sleeping the same amount every time, consider increasing delay.
4. **Reuse a `requests.Session()` (optional but best practice)**
If this function is called repeatedly, passing a session or creating one outside i…
4.4s718 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-5.4 Nano or Inkling?
- Per 1M output tokens, GPT-5.4 Nano is ¥223.38 and Inkling is ¥723.74 on FastMetal (yen, before tax), so GPT-5.4 Nano is cheaper.
- How do the context windows of GPT-5.4 Nano and Inkling compare?
- GPT-5.4 Nano takes 400,000 tokens; Inkling takes 524,288.
- Which ranks higher, GPT-5.4 Nano or Inkling?
- Inkling ranks higher on the public arena (GPT-5.4 Nano #107, Inkling #63). Ranks move as the leaderboard updates.
- Can I use GPT-5.4 Nano and Inkling with the same API key?
- Yes. On FastMetal's OpenAI-compatible endpoint, switch by passing "gpt-5.4-nano" 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-5.4 Nano and Inkling from the browser chat and the API. No monthly fee.