Meta has unveiled Muse Spark 1.3, the newest version of its low-cost reasoning model, with the company claiming its largest improvements yet in coding and agentic tasks. The release, announced late Wednesday, arrives less than a month after Muse Spark 1.2 and is now available through Muse Code and the Meta Model API. CEO Mark Zuckerberg promoted the launch on social media, calling it "frontier performance almost too cheap to meter" and stating it represents the company's biggest leap forward in coding and agentic work to date.
According to Meta's benchmark comparisons, Muse Spark 1.3 achieved 75.4% on the DeepSWE coding benchmark and 98.1% on the 512K-1M version of the MRCR long-context test when run at its new "max" reasoning level. Independent testing by Artificial Analysis gave the publicly available Muse Spark 1.3 xhigh version a score of 61 on its Intelligence Index, four points ahead of Spark 1.2 and equal to GPT-5.6 Sol max, Grok 4.6 high, and Claude Opus 5 high. The limited-preview max version scored 62, placing it behind only Claude Fable 5.1 and Claude Opus 5 among models in the comparison at launch. On cost efficiency, Artificial Analysis calculated that Spark 1.3 xhigh costs approximately $0.55 per Intelligence Index task, the lowest of any model scoring 59 or higher on its index, compared to $0.95 for GPT-5.6 Sol max and $0.94 for Grok 4.6 high.
The report notes that Meta's own benchmark figures were generated through the Meta Model API, while comparison numbers came from a mix of Meta's own evaluations, official leaderboards, and results reported by competing model providers. The company describes its testing of third-party models as "best-effort," meaning the table shouldn't be interpreted as a single independent head-to-head test conducted under identical conditions. Zuckerberg also confirmed that open-weight releases of Muse Spark will arrive "coming soon," a move that could allow developers to download and run the model on their own infrastructure, bypassing Meta's hosted API, though the company hasn't yet published the license terms that will accompany the release.
The timing of Spark 1.3's release positioned it directly against Google's Gemini 3.8 Flash, which launched just hours earlier with its own claims of improved reasoning and coding at $0.75 per million input tokens. Artificial Analysis initially placed Gemini 3.8 Flash (high) on its Intelligence-versus-Cost Pareto frontier with a score of 59 at $0.58 per task, but Muse Spark 1.3 xhigh arrived shortly after at 61 and $0.55 per task, pushing Google's model off that frontier. The report highlights that Muse Spark 1.3 is the fourth version of the model Meta has shipped since April, following Muse Spark 1.1 in July and 1.2 the month after. The release comes as part of Meta's broader push around Muse Code, its terminal-based coding agent that officially launched on Tuesday with subscription plans starting at just $5 per month and a new SDK entering developer preview. Meta is competing not just on model quality but at the platform level, where Anthropic's Claude Code and OpenAI's Codex have established the standard for how developers actually interact with an agent in daily workflows. The strategic question for enterprise buyers will be whether price advantages alone can overcome the ecosystem gravity already built around incumbent tools, particularly when switching costs in developer workflows run high.