Trying Out Muse Spark 1.3 on OpenCode

ChatGPT Image 2026年9月6日 18_02_20
ChatGPT Image 2026年9月6日 18_02_20

The much-talked-about Muse Spark 1.3 became available on OpenCode, so I tried it out immediately.

Price (Comparison with DeepSeek V4)

Compared to the most widely used DeepSeek V4 Flash on OpenCode, the pricing looks like this.
(DeepSeek V4 has different prices during peak and off-peak times, so both are listed here)

Item Muse Spark 1.3 Contributor DeepSeek V4 Flash Off-Peak DeepSeek V4 Flash Peak
Input $0.10 $0.22 $0.44
Output $0.20 $0.66 $1.32
Cached read $0.002 $0.007 $0.014
Cached write — — —
Go Monthly Usage Quota $60 equivalent $30 equivalent $30 equivalent
5-hour estimate 45,300 req 7,600 req consumes quickly within the same quota
Weekly estimate 113,300 req 18,900 req same as above
Monthly estimate 226,600 req 37,800 req same as above

It’s quite cheap.

Looking at the price comparison alone, Muse Spark 1.3 Contributor is approximately

  • 2.2 times cheaper for Input
  • 3.3 times cheaper for Output
  • 3.5 times cheaper for Cache

than DeepSeek V4 Flash during off-peak times. Compared to DeepSeek during peak times, the difference is even larger:

  • 4.4 times cheaper for Input
  • 6.6 times cheaper for Output
  • 7 times cheaper for Cache

Also, by selecting Muse Spark 1.3 Contributor Free on OpenCode Zen, you can use it for free with time restrictions. I expected it would hit the rate limit immediately, but it actually worked quite well.

Impressions

The two most impressive aspects were:

  • Intelligent. It feels smart
  • Fast. Responses are quick, which reduces the overall time to complete tasks

This development speed is addictive. I was very impressed.

Using for Input/Output Learning

This is the most important point to be careful about. Muse Spark 1.3 Contributor explicitly states that it uses user input and output for training. Therefore, it’s safer not to use it for confidential information, personal data, sensitive data, authentication information, software with unique logic, or software that performs special authentication.

Of course, confidential information, personal data, or sensitive data wouldn’t be included in source code anyway, so it’s not a problem now. However, there’s no guarantee it won’t accidentally read files outside the project, so using it in a sandbox environment like Docker Sandboxes would be safer.

Should I Use It?

Given the above, there are risks to using it at work, but it should be fine for personal programs or open-source projects. I plan to use it for now, even though there’s a possibility of price increases in the future.

For now, my usage plan will be:

  • Use Muse Spark 1.3 Contributor if it’s okay for source code to be used for training
  • Use DeepSeek V4 Flash if source code shouldn’t be used for training

That’s the general approach.

© 2025 Hiroe Tech Notes. All rights reserved.

Comments

No comments yet.