Decision summary
- Muse Code or Meta Model API -> muse-spark-1.3
- Self-hosted weights -> Muse Spark Open
- Discounted hosted calls -> muse-spark-1.3-contributor, not the open weights
Muse family comparison
Frontier comparisonMuse Spark Open vs Muse Spark 1.3
Muse Spark Open is Meta's October 15, 2026 open-weight release: an 816B/42B mixture-of-experts model with a 1 million token context under the Muse Model License.
Featured · Updated today · Last verified: October 2026
Choose Muse Spark Open when
Teams that can host weights and accept the Muse Model License.
Choose Muse Spark 1.3 when
Hosted Muse coding and agent calls without operating the weights.
- Muse Code or Meta Model API: muse-spark-1.3
- Self-hosted weights: Muse Spark Open
- Discounted hosted calls: muse-spark-1.3-contributor, not the open weights
Decision axes: Best for · Access · Context · List price
Reasoning
Hosted Spark 1.3 documents effort including max. Open weights leave the runtime to you.
Coding
1.3 is the hosted coding API. Open is the weight release.
Multimodal
Open accepts text, image, audio, and video. Hosted 1.3 accepts text, image, video, and PDF.
Speed
You own Open's latency. Meta owns 1.3's.
Enterprise
The Muse Model License is not the hosted terms. Contributor mode applies to the API, not the weight files.
Curated matrix for Muse Spark Open vs Muse Spark 1.3 — confirm live limits and pricing on each vendor’s official pages.
How they compare
Criterion-by-criterion notes from the catalog—not a ranking. Validate on your own gold set.
| Criterion | Muse Spark Open | Muse Spark 1.3 |
|---|---|---|
| Best for | Teams that can host weights and accept the Muse Model License. | Hosted Muse coding and agent calls without operating the weights. |
| Access | Weights on Hugging Face as meta-models/Muse-Spark-Open-816B-A42B, plus an NVFP4 variant. Released October 15, 2026. Not an API ID. | API ID muse-spark-1.3 in Muse Code and the Meta Model API. Released September 2, 2026. |
| Context | 1 million tokens. Text, image, audio, and video in. Text out. | 1,048,576 tokens. Text, image, video, and PDF in. Text out. |
| List price | No hosted token price. You pay for your own GPUs. The license has limited commercial terms. | Standard mode on the Meta pricing page: $1.25 input, $0.15 cached input, $4.25 output per million tokens. Contributor ID muse-spark-1.3-contributor is discounted. |
| Controls | You choose the runtime. There is no Meta contributor-mode switch on the weight files. | Reasoning effort includes max on the Standard tier. |
| Operational risk | Read the license before you ship. This is not muse-spark-1.3 and not a drop-in client change. | Contributor mode lets Meta train on your content. Standard mode does not, and it costs more. |
Key insights
Concrete technical or product signals.
- Open weights and the hosted API are different products that share the Spark name.
- Spark 1.3 remains the current hosted Spark. Spark Open does not supersede it.
Use cases
Where this shines in production.
- Deciding whether to self-host Spark
- Keeping a Muse Code key on muse-spark-1.3
Limitations & trade-offs
What to watch for.
- No quality table between the open weights and the hosted API is scored here.
- License terms can change; read the card that ships with the weights.
Who should not choose this?
Signals that this pair is the wrong default.
- Do not point an API client at the Hugging Face repo name.
- Do not treat Spark Open as the new hosted default.
- Do not skip the license.
Final recommendation
What to do after reading the table.
Keep production on Spark 1.3 unless you have a reason to own the weights.
FAQ
Is Muse Spark Open better than Muse Spark 1.3?
Keep production on Spark 1.3 unless you have a reason to own the weights.
Which is cheaper: Muse Spark Open or Muse Spark 1.3?
Listed unit prices: Muse Spark Open No hosted token price. You pay for your own GPUs. The license has limited commercial terms; Muse Spark 1.3 Standard mode on the Meta pricing page: $1.25 input, $0.15 cached input, $4.25 output per million tokens. Contributor ID muse-spark-1.3-con. A lower listed unit price is not the same as a lower total workload cost—tokens, session minutes, caching, introductory rates, and add-on modes can reverse the ranking. Confirm live rates on each vendor’s official page.
Can I use both Muse Spark Open and Muse Spark 1.3?
Yes. Many teams keep both and route by constraint. Use Muse Spark Open for Teams that can host weights and accept the Muse Model License, and Muse Spark 1.3 for Hosted Muse coding and agent calls without operating the weights. Confirm vendor terms before mixing them in one production workflow.