GPT-5.6 Sol vs Mercury 2.5
Pricing, context window and real answers (September 2026)
GPT-5.6 Sol (OpenAI) and Mercury 2.5 (Inception), 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-5.6 Sol | Mercury 2.5 | |
|---|---|---|
| Provider | OpenAI | Inception |
| Input (per 1M tokens) | ¥893.5 | ¥6.77 |
| Output (per 1M tokens) | ¥5,361 | ¥25.38 |
| Typical cost (1,000 in + 500 out tokens × 1,000 calls) | ¥3,574 | ¥19 |
| Context window | 1,050,000 tokens | 260,000 tokens |
| Release date | 7/9/2026 | 9/8/2026 |
| Input modalities | file, image, text | text |
| Arena · overall | #17 · ELO 1,483 | Unranked |
| Arena · Japanese | #2 · ELO 1,522 | Unranked |
| Arena · coding | #14 · ELO 1,529 | Unranked |
Prices are FastMetal's yen rates before tax. Arena ranks are as of the latest leaderboard scrape.
Which should you pick?
- On price, Mercury 2.5: ¥25.38 per 1M output tokens, 100% less than GPT-5.6 Sol.
- For long documents, GPT-5.6 Sol: a 1,050,000-token context window against 260,000 for Mercury 2.5.
- 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-5.6 Sol
“strawberry” can be split into letters:
s, t, **r**, a, w, b, e, **r**, **r**, y
The letter **r** appears **3 times**.
3.0s78 output tokens
Mercury 2.5
To determine the number of 'r's in the word "strawberry", we can examine each letter in sequence:
1. **s** - Not 'r'
2. **t** - Not 'r'
3. **r** - First 'r' found
4. **a** - Not 'r'
5. **w** - Not 'r'
6. **b** - Not 'r'
7. **e** - Not 'r'
8. **r** - Second 'r' found
9. **r** - Third 'r' found
10. **y** - Not 'r'
Counting these occurrences, there are **3** 'r's in "strawberry".
1.8s968 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-5.6 Sol
The error occurs because `users` is `undefined` when `UserList` renders, so JavaScript cannot call `.map()` on it. This often happens when data is loaded asynchronously or the parent does not pass the prop.
### Fix options
**1. Give `users` a default value:**
```jsx
function UserList({ users = [] }) {
return (
<ul>
{users.map(user => (
<li key={user.id}>{user.name}</li>
))}
</ul>
);
}
```
Note that this handles `undefined`, but not an explicitly passed `null`.
**2. Handle missing or invalid data explicitly:**
```jsx
function UserList({ users }) {
if (!Array.isArray(users)) {
return <p>No users available.</p>;
}
return (
<ul>
{users.m…
7.1s395 output tokens
Mercury 2.5
This error occurs because the `users` prop is `undefined` when the component renders. Since the code calls `.map()` on `users`, JavaScript throws a TypeError when `users` is not an array.
Here is how to fix it:
1. **Set a Default Value**
Update the function signature to default `users` to an empty array if it is not provided:
```javascript
function UserList({ users = [] }) {
// ...
}
```
2. **Add a Conditional Check**
If `users` might be null or undefined during data loading, guard the map operation:
```javascript
{users && users.map(user => <li key={user.id}>{user.name}</li>)}
```
3. **Verify Parent Props**
Ensure the parent component pas…
2.4s1028 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-5.6 Sol
### Issues and suggested improvements
- Move imports to module scope.
- Avoid bare `except:`; it catches programming errors and interrupts such as `KeyboardInterrupt`.
- Use `response.raise_for_status()` instead of manually checking only for `200`; other 2xx responses may also be valid.
- Retry only transient failures, such as timeouts, connection errors, `429`, and selected `5xx` responses. Retrying most `4xx` errors is not useful.
- Use exponential backoff rather than a fixed delay.
- Avoid sleeping after the final failed attempt.
- Replace `item["active"] == True` with `item.get("active") is True`.
- Use a list comprehension for the transformation.
- Add type hints and validate arguments…
33.1s1646 output tokens
Mercury 2.5
Here is a review of the function with suggestions for improvement.
### Key Observations
1. **Imports:** Libraries should be imported at the module level, not inside functions.
2. **Exception Handling:** Bare `except:` clauses catch system exits and interrupts. Use specific exceptions like `requests.exceptions.RequestException`.
3. **Logic & Pythonicity:**
* Avoid `== True`; rely on truthiness or `item.get("active")`.
* List comprehensions are more concise and efficient than manual appending.
4. **Retry Strategy:** Implement exponential backoff instead of fixed sleeps, and avoid sleeping after the final failed attempt.
5. **Documentation:** Add type hints and a docstring fo…
4.6s1999 output tokens
Frequently asked questions
- Which is cheaper, GPT-5.6 Sol or Mercury 2.5?
- Per 1M output tokens, GPT-5.6 Sol is ¥5,361 and Mercury 2.5 is ¥25.38 on FastMetal (yen, before tax), so Mercury 2.5 is cheaper.
- How do the context windows of GPT-5.6 Sol and Mercury 2.5 compare?
- GPT-5.6 Sol takes 1,050,000 tokens; Mercury 2.5 takes 260,000.
- Can I use GPT-5.6 Sol and Mercury 2.5 with the same API key?
- Yes. On FastMetal's OpenAI-compatible endpoint, switch by passing "gpt-5.6-sol" or "mercury-2.5" 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-5.6 Sol and Mercury 2.5 from the browser chat and the API. No monthly fee.