Meta's Muse Spark jumps to 1.3 — same price tag.
Last month's 1.1 was priced right but clearly a step behind the top models. Version 1.3 keeps the $1.25/$4.25 rate unchanged while lifting output quality, which makes this a bigger win for cost-conscious developers than for casual users.
Last month's 1.1 was
priced right, ranked behind
Muse Spark sits alongside Glimmer and Llama 4 as one of Meta AI's core models, handling reasoning and agentic tool use. Last month's 1.1 release earned praise on pricing, but its output quality still trailed the top-tier models. Version 1.3 skips straight past 1.2 and holds the $1.25/$4.25 (input/output, per 1M tokens) rate unchanged while raising performance on its own.
Improving performance without touching price isn't unheard of, but it amounts to a de facto price cut. The official Meta AI blog frames this update as primarily about cost efficiency.
Muse Spark's main job is agentic work — calling external tools and APIs to get a task done — rather than plain text generation. Since tool-call accuracy is often the real bottleneck in production, a jump in success rate like this one is the kind of improvement that's easy to trace straight to real-world impact. Full specs are published on Meta's model page.
| Muse Spark 1.1 | Muse Spark 1.3 |
|---|---|
| Shipped August 2026 | Shipped September 2026 |
| $1.25 / $4.25 (input/output) | $1.25 / $4.25 (unchanged) |
| Rated a step behind top models | Higher tool-use success rate |
| 128K token context window | 128K token context window (unchanged) |
Same price, better underneath.
The performance gain,
in numbers
Price holds steady while tool-use success rate rises measurably.
Who benefits, and how
Engineers
With tool-use success climbing from 71% to 79%, it's more reasonable to route agentic workflows that call external APIs through Muse Spark. Since pricing hasn't moved, existing cost models carry over as-is. If a workflow was previously routed to a pricier top-tier model just because Muse Spark's success rate was too low, this update might let you route it back and cut costs.
Business / Back Office
For cost-focused teams, this is close to a free upgrade — better output, no price increase. Lighter workloads where the gap is barely noticeable may not see the benefit at all. Not having to rebudget is a small but genuinely useful win for whoever owns the cost model.
Re-test existing agent flows on 1.3
If you run automation that depends on tool calls, compare its success rate against the 1.1-era baseline to confirm the gain.
Leave your cost model as-is
Since pricing hasn't changed, there's no need to rebudget — treat the performance gain as pure upside.
Not all upside
The jump from 71% to 79% comes from Artificial Analysis, a third-party benchmark — not an internal Meta figure, and not something AI Navigate independently verified. Benchmarks measure a specific slate of tool-use tasks; real workflows may show a smaller or larger gap.
Meta also hasn't fully explained what changed internally to justify skipping straight from 1.1 to 1.3 without a 1.2 release. Skipping version numbers isn't unusual, but it does leave something to be desired on the transparency front. Meta also hasn't published much detail on the kind or difficulty of the tasks used to measure that success rate, so it's worth testing against your own use case rather than taking the number at face value.