Muse Spark 1.3
Meta's fast flagship in this snapshot. Strong, not the leader, and one of the quickest models that still sits in the mid-40s.
No High score is published. Showing Extra high.
Every published effort
| Setting | Index | Per task | Tokens / sec |
|---|---|---|---|
| Max | 48 | $1.60 | 181 |
| Extra high | 45 | $1.37 | 142 |
What this workload costs
List price for one call. Reasoning tokens count as output. A cache hit uses the lab’s cache rate when one is published.
Specs
- API id
- Not published here
- Released
- Sep 2026
- Context
- 1.05M
- Max output
- 944k
- Input / 1M
- $1.25
- Output / 1M
- $4.25
- Cache read / 1M
- $0.15
- Knowledge
- —
- Inputs
- Text, Image, Audio, Video
- Weights
- Closed
Standard tier as reported on 3 Sep 2026. A cheaper contributor tier was also reported ($0.10 / $0.20) and is not the price used here. Re-check Meta before a large deployment. The release day is not more precise than September 2026 in the sources used here. It was already on the index by 7 Sep.
Modalities and the token price are from a 3 Sep 2026 write-up of Meta's pricing page, not re-read from Meta on 1 Oct.
Use it when
Multimodal product work where speed matters. Max effort is 48 on the index at about 181 tokens per second. Extra-high is 45 and a bit cheaper per task.
Leave it when
Assuming the weights are open. Muse Spark is treated as a closed API model here. Llama-style open checkpoints are a different card, and they are not this score.