Muse Spark 1.3

The TrueFoundry article, published September 22, 2026, summarizes Meta Superintelligence Labs' release of Muse Spark 1.3 on September 2, 2026, about four weeks after 1.2. It is framed as a natively multimodal model targeted at agentic and coding work, available through Muse Code and the Meta Model API. The article note Key changes emphasized are efficiency rather than headline intelligence: roughly 20% fewer tool calls and 25% fewer tokens than 1.2 per Meta's own engineers, alongside a max reasoning level, better multitasking inside one long thread, and stronger resistance to prompt injection. Weights remain closed, with open weights

Meta Superintelligence Labs released Muse Spark 1.3 as an agentic coding model, with the MarkTechPost coverage published on September 3, 2026. The model is described as the fourth Muse Spark release in five months and is positioned for long-horizon agentic and coding work rather than single-turn generation. According t The candidate highlights efficiency gains as the core change: Meta's engineers measured roughly 20% fewer tool calls and 25% fewer tokens versus Muse Spark 1.2 on comparable workloads. Training spanned multiple agent harnesses so the behavior generalizes beyond a single environment, and the model is designed to hold se

The Flowtivity article, published September 2, 2026, covers Meta Muse Spark 1.3's release and its benchmark positioning for AI agents. It reports 75.4% on DeepSWE 1.1 for end-to-end agentic software engineering, 88.8% on Terminal-Bench 2.1, 59.4% on SWEAtlas CodeBase QnA, and 98.5% on long-context retrieval, all within Availability was same-day in Muse Code and the Meta Model API, with paid developer access opening September 3. The piece frames Muse Spark 1.3 as Meta's first closed, directly monetized frontier line, shipping on a four-week cadence and pressuring rivals on price. It also notes Claude Opus 5 still leads Meta's own tabl

The Axios piece, dated September 2, 2026, reports Meta's release of Muse Spark 1.3, describing it as a model update that significantly improves coding and agentic performance. The coverage places the launch in the context of a week in which Anthropic, OpenAI, and Google also made major model announcements, framing Meta Meta AI chief Alexandr Wang told Axios the update helps pave the way for upcoming products such as personal agents, calling the model very competitive with frontier models and highlighting usability improvements tied to features Mark Zuckerberg has discussed on earnings calls, including personal agents. The excerpt is

The LLM-Stats page for Muse Spark 1.3, dated September 2, 2026, lists benchmark scores and pricing data attributed to research.meta.ai and Meta's official blog or scorecard. Reported scores include a GDPval-AA Elo of 1754.00/3000 (rank 4), MRCR v2 (8-needle) at 0.98/1 in the 256K–512K context range (rank 1), and an MRC The page positions Muse Spark 1.3 within a cost-efficiency leaderboard, showing a blended price per million tokens of $1.39 versus competing models such as Gemma 4 E4B, GPT OSS 120B, DeepSeek-V4-Flash, DeepSeek-V4.1-Flash, and Claude Opus 5.5. Performance by conversation depth and a quality tracker are also surfaced, w