Muse Spark 1.3 Contributor

This DataCamp tutorial explicitly names "muse-spark-1.3-contributor" as the default model that Muse Code starts users on, and confirms the model was released September 2, 2026 by Meta Superintelligence Labs with a 1,048,576-token context window and multimodal input (text, images, video, files). It documents that the co Hands-on testing on three coding tasks across six runs found that 1.3 was cheaper on two tasks and 38% more expensive on the third, for a net 12% cost increase versus 1.2. Model completions fell 23% and 32% on the two tasks 1.3 won, broadly aligning with Meta's claim of fewer tool calls, though uncached input token red

Coursiv covers Meta's September 2, 2026 release of Muse Spark 1.3, describing it as a proprietary, closed-weights multimodal reasoning model accepting text, images, and video across a roughly one-million-token context. It documents the contributor-tier pricing at $0.10 per million input tokens with the trade-off that M The article positions Muse Spark 1.3 for long-running, multi-agent workflows where the model must track context across steps, handle conflicting inputs, and ask clarifying questions rather than guess. Independent testing is reported as tying the model to GPT-5.6 Sol, Grok 4.6, and Claude Opus 5 on Artificial Analysis's

explainx.ai covers the Muse Spark 1.3 launch, which it dates to September 3, 2026 as the replacement for Muse Spark 1.2 inside Muse Code. It reports Meta's own benchmark table showing Muse Spark 1.3 at 75.4 on DeepSWE v1.1 (ahead of Claude Opus 5's 74.0 and GPT-5.6 Sol's 73.0) and tying GPT-5.6 Sol at 88.8 on Terminal- The piece documents the contributor-tier pricing structure as up to 21x cheaper than standard in exchange for letting Meta train on user usage. It notes the model leads or ties Opus 5 on 2 of 9 published benchmarks (DeepSWE v1.1 and Terminal-Bench 2.1), while Opus 5 still leads on GDPVal-AA v2, JobBench, OSWorld 2.0, D

Meta released Muse Spark 1.3 on September 2, 2026 — its fourth Muse Spark release since April — and shipped it the same day in Muse Code (its terminal-based coding agent) and through the Meta Model API. The page documents two pricing tiers: a standard rate of $1.25 per million input tokens and $4.25 per million output The article separates Meta's vendor-run benchmark numbers from independent reads, noting that the headline benchmark column uses "max" reasoning, which Meta says is not publicly shipped yet pending additional safety testing. On Meta's own table, 1.5 is presented as level with Claude Opus 5 on agentic and professional w