>_TheQuery
← All Comparisons

Muse Spark 1.2 vs Muse Spark 1.3

Muse Spark 1.2Muse Spark 1.3
Pricing
Standard per 1M$1.25 / $4.25$1.25 / $4.25
Cache reads per 1M$0.15$0.15
Contributor per 1M$0.10 / $0.20$0.10 / $0.20

Nothing moved - including the Contributor data contract (details).

Muse Spark 1.2Muse Spark 1.3
Coding
DeepSWE v1.155.0%*75.4%*
Terminal-Bench 2.182.9%†88.8%

Meta ran 1.3 at max reasoning against 1.2 at xhigh. †1.2 figure from its own August chart at a different setting.

Twenty points on DeepSWE, six on Terminal-Bench - with part of both gaps coming from the higher effort setting.

Muse Spark 1.2Muse Spark 1.3
Efficiency
Tool callsBaselineAbout 20% fewer
Tokens usedBaselineAbout 25% fewer

Meta-measured; note the counter-reading that 1.3 (max) used 120M tokens on the AA index against a 72M class median.

Muse Spark 1.2Muse Spark 1.3
Long context
MRCR 256K-512K66.3%98.5%
MRCR 512K-1M55.5%98.1%

Retrieval roughly doubles at both spans - the widest gaps in the comparison.

Muse Spark 1.2Muse Spark 1.3
Knowledge and agents
GDPval-AA v21,631 Elo1,754 Elo
AA Intelligence Index54‡62§

‡Later revised to 57 after an index methodology update. §Max tier, limited partner preview.

Up 123 Elo on knowledge work; the 62 describes a tier most developers cannot call yet.

Muse Spark 1.2Muse Spark 1.3
Availability
API and Muse CodeYesYes
Max reasoning tier
Limited partner preview

Same surfaces; headline 1.3 numbers describe a tier still behind staged access.

Sources

  1. Introducing Muse Spark 1.3
  2. Muse Spark 1.3 - API Pricing and Providers
  3. Introducing Muse Code and Muse Spark 1.2
  4. Muse Spark 1.2: Independent Benchmarks and Analysis
  5. Muse Spark 1.3