Muse Spark 1.3

Muse Spark 1.3 landed in Muse Code and the Meta Model API on September 2, 2026 with every price held from 1.2: $1.25 per million input tokens, $4.25 output, $0.15 cached input on the Standard tier, and $0.10 and $0.20 on the Contributor tier in exchange for letting Meta train on your prompts and completions. Because nothing moved on the rate card, the interesting claim is about consumption rather than price. Meta says 1.3 completes the same agentic work with roughly 20 percent fewer tool calls and 25 percent fewer tokens than 1.2, which is a discount that shows up on the invoice without appearing on the price list, and it is the kind of claim worth measuring on your own traces before you bank it. Reported scores are DeepSWE 1.1 at 75.4, SWEAtlas CodeBase QnA at 59.4, Terminal-Bench tied with GPT-5.6 Sol at 88.8, and MRCR at 98.5 in the 256K to 512K band and 98.1 from 512K to 1M, against 91.5 and 73.8 for Sol. One caveat travels with the comparison table: the 1.3 column runs at max reasoning while the 1.2 column runs at xhigh, so part of the version-over-version jump is a reasoning-tier gap rather than a model gain. Context is 1 million tokens. Despite what the Muse Spark name has come to imply, 1.3 is not an open-weight model. Meta says a Muse Spark open-weight release is coming and has named no date, no variant, and no license.