Meta released Muse Spark 1.3 on Wednesday, calling it the company's most capable model and claiming it matches or exceeds OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5.1 on coding and agentic benchmarks4,10. CEO Mark Zuckerberg described the release as Meta's "biggest jump" yet in coding and agentic work3. Chief AI Officer Alexandr Wang said the model surpasses GPT-5.6 Sol in coding capability and matches Claude Fable 5.1.
The release carries a material caveat. Meta's headline benchmark results come from a max reasoning configuration that is still completing additional safety testing and is not broadly available to developers. Artificial Analysis, the third-party benchmarking firm, said it evaluated max in a limited partner preview and currently lists no API provider for that configuration. The version rolling out this week through Muse Code and the Meta Model API uses the xhigh reasoning setting.
The gap between the two configurations is measurable. Meta reports GDPval-AA v2 scores of 1,754 Elo for max versus 1,709 for xhigh, and OSWorld 2.0 scores of 66.9 versus 57.2. Artificial Analysis scored max at 62 on its Intelligence Index and xhigh at 612. On Terminal-Bench 2.1, xhigh scored 89.2, up from Muse Spark 1.2's 82.9.
ANALYSIS The distinction matters because the model developers can actually deploy today is xhigh, not the max configuration that anchors Meta's competitive claims against Opus 5 and GPT-5.6 Sol.
Even at xhigh, the model is competitive on price-performance. Artificial Analysis said Muse Spark 1.3 xhigh costs approximately $0.55 per Intelligence Index task, compared with $0.95 for GPT-5.6 Sol max and $0.94 for Grok 4.6 high. Meta kept Standard pricing unchanged at $1.25 per million input tokens and $4.25 per million output tokens. The Contributor tier, which grants Meta permission to use prompts and completions for training, remains at $0.10 per million input tokens and $0.20 per million output tokens, a discount exceeding 90%7.
Meta's internal engineering comparisons found Muse Spark 1.3 used roughly 20% fewer tool calls and approximately 25% fewer tokens than Muse Spark 1.2 during coding work5,6. Artificial Analysis measured xhigh throughput at 235.2 output tokens per second.
The release arrived the same day Google released Gemini 3.8 Flash, which Google called its best reasoning and coding Flash model yet. Artificial Analysis scored Gemini 3.8 Flash at 59 on its Intelligence Index at $0.58 per task, placing it below Muse Spark 1.3 xhigh on both measures. Wang responded to the Artificial Analysis report on X, writing "gemini who?".
Muse Spark 1.3 is the fourth version of the model Meta has shipped since April, arriving less than a month after Muse Spark 1.2. Muse Code, Meta's coding agent, formally launched out of beta on Tuesday with three subscription plans starting at $5 per month and a new SDK in developer preview.
Zuckerberg said on X that "Muse Spark open weights releases" are coming soon. Muse Spark 1.3 itself shipped as a proprietary model. Meta's blog post references "the Muse Spark open weights release" without identifying a version, release date, model size, or license. Wang confirmed that development of Meta's next model, codenamed Watermelon, is progressing as planned.
Meta's API platform has seen strong adoption, with some developers using trillions of tokens per week. The company raised its 2026 capital expenditure guidance to $130 billion to $145 billion, nearly double the roughly $72.2 billion spent in 2025.