Summary
- Meta has released Muse Spark 1.2, which scored 54 on the Intelligence Index tracked by evaluation firm Artificial Analysis.
- The model marks Meta's third release in four months, and was noted for a significant improvement in agentic knowledge work capabilities compared to the previous version.
- The score of 54 reportedly places it in a tie for third place alongside SpaceXAI, though detailed breakdowns by benchmark category were not disclosed.
Meta unveils Muse Spark 1.2
Meta has released a new model, Muse Spark 1.2. Evaluation firm Artificial Analysis reported that the model scored 54 on its Intelligence Index. The score matches that of SpaceXAI, putting the two in a tie for third place, according to the firm. Which models occupy first and second place, and the exact score gaps between them, were not included in this disclosure.
Third release in four months
What stands out is the pace of release. According to Artificial Analysis, Muse Spark 1.2 is Meta's third model in four months. The pattern of rapid version updates aimed at boosting scores appears to reflect an effort to close the gap with the top-ranked group.
| Item | Confirmed details |
|---|---|
| Model | Muse Spark 1.2 |
| Intelligence Index | 54 |
| Ranking | Tied for third (with SpaceXAI) |
| Release cadence | 3 releases in 4 months |
| Key improvement | Agentic knowledge work |
Improvement focus: "agentic knowledge work"
The evaluation firm noted a clear improvement in agentic task performance compared to previous releases. Artificial Analysis described it as "significantly improving agentic knowledge work capabilities." This is interpreted to mean gains in multi-step tasks involving documents, search, and tool use, rather than single-turn question answering. However, no detailed figures were given on which specific sub-categories improved or by how much.
Trends in per-model Intelligence Index scores and shifts in how agentic performance is evaluated can be tracked further in METAL LAB's benchmark coverage.
What remains unconfirmed
The information currently available is based on a single post from the evaluation firm. Technical specifications—including whether model weights will be released, context length, pricing policy, and training/inference infrastructure—have not been confirmed. The specific model name for SpaceXAI's tied entry was also not disclosed.
| Category | Status |
|---|---|
| Intelligence Index score | Confirmed |
| Ranking | Confirmed (tied for third) |
| Detailed benchmark breakdown | Not disclosed |
| Model specs / release format | Not disclosed |
| Top 1st/2nd place models | Not mentioned |
Further benchmark results and an official announcement from Meta are expected to provide a clearer picture of the score's significance and the competitive landscape.
Comments