Meta shipped Muse Spark 1.3 on 3 September in two configurations. The broadly available xhigh scores 61 on the Intelligence Index, while the stronger max is still in safety testing and has not reached developers. Despite earlier commitments the model came out closed, with no open weights.
Key takeaways
- Two configurations: xhigh broadly available, max still in safety testing
- Intelligence Index: 61 points for xhigh — level with GPT-5.6 Sol max and Claude Opus 5 high
- Claude Fable 5.1 leads on 66 points
- Pricing unchanged: $1.25 per million input tokens and $4.25 per million output
- The cheaper Contributor tier costs $0.10 and $0.20, but lets Meta use the data for training
Frontier in a version you cannot get
Meta showed two variants of the same model and attributed its best results to the one that stays unavailable. The xhigh configuration, the deployable one, stops at 61 points on the Intelligence Index — the same as GPT-5.6 Sol in max mode and Claude Opus 5 in high mode. The max configuration is still in safety testing, and Meta promises access later without naming a date.
| Model / configuration | Intelligence Index |
|---|---|
| Claude Fable 5.1 | 66 |
| Muse Spark 1.3 xhigh | 61 |
| GPT-5.6 Sol max | 61 |
| Claude Opus 5 high | 61 |
Price per result, not per token
More interesting than the score is the economics. Pricing has not moved since 1.2. On coding tasks the model trades wins with rivals from OpenAI and Anthropic, improving clearly over the previous version.
| Tier | Input / M tokens | Output / M tokens | Data used for training |
|---|---|---|---|
| Standard | $1.25 | $4.25 | no |
| Contributor | $0.10 | $0.20 | yes |
Contributor, or paying in data
The Contributor tier costs less than a tenth of the standard rate. The difference is that in this mode Meta may use the traffic for training. It puts the trade plainly: you pay in money or in your users' tokens.
Why it matters
Publishing results from a configuration developers cannot reach blurs the line between what a model can do and what you can buy. Separately, dropping open weights matters in its own right — Meta was the loudest advocate of openness among the large labs, and another closed launch shifts that reference point.
What's next
- Meta says the max configuration will be released once safety testing finishes, with no date given
- The absence of open weights in 1.3 leaves the roadmap unclear for teams counting on downloadable Muse Spark models
Sources
- VentureBeat — Meta says Muse Spark 1.3 has frontier performance — but its best results come from a model developers can't broadly use yet
- Artificial Analysis — Intelligence Index
