By admin 
Meta unveiled its latest artificial intelligence model, Muse Spark 1.3, yesterday, marking a significant leap in performance and efficiency compared to its predecessor. According to Meta’s official announcement and subsequent commentary, the new model exhibits faster processing speeds and superior performance on third-party benchmarks. This advancement has been heralded by Meta’s leadership, with co-founder and CEO Mark Zuckerberg describing it on X as Meta’s "biggest jump yet" in capabilities for coding and agentic tasks, even coining the phrase "frontier performance almost too cheap to meter."
The substance behind these claims is considerable. Muse Spark 1.3 demonstrates substantial improvements over the version released just last month, Muse Spark 1.2, particularly in handling long-running agent tasks. The version now accessible to developers is positioned as one of the most cost-effective, high-performance options available, rivaling top-tier independent model rankings. Meta highlights that its most impressive benchmark results for Muse Spark 1.3 are derived from its "max reasoning configuration." However, this top-tier version is currently undergoing additional safety testing and is slated for release "shortly." Independent benchmarking firm Artificial Analysis, which evaluated the "max" configuration under a limited partner preview, currently lists no API provider for it, underscoring its restricted availability.
The version that is broadly rolling out this week, accessible through Meta’s Muse Code harness and the Meta Model API, utilizes Meta’s previously established reasoning settings, including the "xhigh" configuration. This distinction is crucial for enterprises evaluating the practical deployment of Muse Spark 1.3. The pertinent question for businesses is not whether Muse Spark 1.3 can reach frontier territory, but rather how close the currently deployable version gets to that benchmark and at what real-world cost.
The Shipping Model: Strong Performance, Not Yet the Absolute Benchmark Leader
Meta has been transparent about the performance metrics of both configurations, disclosing results for both in its underlying evaluation reports. This means the company is not concealing the capabilities of the deployable model. However, its launch materials prominently feature the "max" variant, and some of the most impressive benchmark scores are attributed to this configuration. For instance, Meta reports GDPval-AA v2 scores of 1,754 Elo for the "max" version compared to 1,709 for "xhigh." Similarly, OSWorld 2.0 scores are 66.9 for "max" versus 57.2 for "xhigh," and JobBench scores stand at 64.9 for "max" against 61.2 for "xhigh."
Interestingly, on certain tests, the performance difference is negligible or even favors the "xhigh" configuration. DeepSearchQA, for example, is tied at 89.4, while the "xhigh" version scores 89.2 on Terminal-Bench 2.1, slightly edging out the "max" configuration’s 88.8.
Independent analysis from Artificial Analysis places Muse Spark 1.3 "max" at 62 on its Intelligence Index, with the widely available "xhigh" version scoring 61. This latter score places it in direct contention with models like GPT-5.6 Sol max, Grok 4.6 high, and Claude Opus 5 high, all of which achieved a score of 61. However, Anthropic’s models still hold the top positions on the leaderboard. Claude Fable 5.1 reaches 66 at its "max" setting and 65 at "xhigh," while Claude Opus 5 achieves 63 at both "max" and "xhigh." In essence, Muse Spark 1.3 "xhigh" is firmly within the frontier cluster of AI models, but it is not currently the model setting the absolute pace at the very top.
This represents a substantial improvement from Muse Spark 1.2. Previous coverage highlighted Meta’s entry into the coding wars with a credible challenger that generally lagged behind Anthropic’s leading models. Muse Spark 1.2 scored 82.9% on Terminal-Bench 2.1, compared to Opus 5’s 86.7%, and also fell behind Opus on other key coding comparisons presented by Meta. With the 1.3 release, Meta is no longer just participating in the competition; it is actively trading wins with industry giants like OpenAI and Anthropic across several coding and agentic evaluations.
Meta also emphasizes that the underlying model has become more user-friendly. Muse Spark 1.3 is engineered to manage multiple workflows within extended conversations, efficiently gather context using tools, identify gaps in its own planning, solicit user clarification when necessary, and seek confirmation before executing significant actions. In internal comparisons conducted by Meta engineers, Muse Spark 1.3 utilized approximately 20% fewer tool calls and 25% fewer tokens than its predecessor during coding tasks. For enterprises that rely on thousands or even millions of agent interactions, these operational efficiencies could prove more impactful than a marginal increase in leaderboard rankings.
"Almost Too Cheap to Meter" – A Matter of Value, Not Price Cuts
Despite the significant performance gains and Zuckerberg’s enthusiastic "almost too cheap to meter" pronouncement, Muse Spark 1.3 has not seen a reduction in its API pricing. Meta has maintained the same pricing structure for Muse Spark 1.3 as for Muse Spark 1.2: $1.25 per million input tokens, $4.25 per million output tokens, and $0.15 per million cached input tokens. This indicates that Zuckerberg’s statement is less about a decrease in per-token costs and more about the enhanced value and capabilities that developers can achieve with those tokens.
The accompanying table details the pricing of various AI models, illustrating Meta’s positioning. Muse Spark 1.2 and 1.3 Contributor models are priced at $0.10 for input tokens and $0.20 for output tokens, totaling $0.30 per million tokens. This tier offers exceptionally low pricing in exchange for Meta’s permission to use prompts and completions for training purposes. While attractive for prototyping, this arrangement presents a distinct data governance consideration for enterprises handling proprietary or sensitive information.
For the standard Muse Spark 1.1, 1.2, and 1.3 models, Meta charges $1.25 per million input tokens, $4.25 per million output tokens, and a total of $5.50 per million tokens. This places it in the mid-to-higher tier compared to some competitors, but the value proposition, as Meta emphasizes, lies in the model’s efficiency and advanced reasoning.
Artificial Analysis provides supporting evidence for Meta’s value claim, though with a caveat. The firm measures Muse Spark 1.3 "xhigh" at an impressive 235.2 output tokens per second, estimating a cost of $0.55 per Intelligence Index task. With an Intelligence Index score of 61, this translates to the lowest cost per task among all models evaluated at that intelligence level. This is a compelling metric for cost-conscious developers.
However, a comparison with Muse Spark 1.2 reveals a nuance. Muse Spark 1.2 cost $0.40 per Artificial Analysis task while scoring 57. This means that despite unchanged per-token pricing, the cost to complete an average task, as measured by Artificial Analysis, has actually increased from generation to generation. This increase is primarily attributed to heavier input-token consumption observed on agentic evaluations, a direct consequence of the model’s enhanced capabilities in complex, multi-turn tasks. While this doesn’t directly contradict Meta’s claim of reduced token usage in specific coding workflows, it highlights the complexity of "cheapness" in AI agents, where token rates, reasoning effort, interaction turns, tool calls, and retries all contribute to the final cost.
Wang’s "Gemini Who?" Remark Lands on a Tight Comparison
Meta’s Chief AI Officer, Alexandr Wang, offered a more pointed assessment of Muse Spark 1.3’s release. Following Artificial Analysis’s publication of the Muse Spark results, Wang reposted them on X with the provocative caption: "i really hate to say it, but… gemini who? 🙄🗿". This remark was particularly pointed given that Google released its Gemini 3.8 Flash model on the very same day. Google positioned Gemini 3.8 Flash for a similar class of workloads, including long-horizon software engineering, autonomous agents, and multi-step professional reasoning. Google touts 3.8 as its best-performing reasoning and coding Flash model to date, marking its third Flash release in just six weeks.
Independent benchmark data provides Wang with some ammunition, though not a decisive victory. Artificial Analysis scores Muse Spark 1.3 "xhigh" at 61 on the Intelligence Index with a cost of $0.55 per task. This compares to Gemini 3.8 Flash "high" reasoning, which scored 59 and cost $0.58 per task. Based on these metrics, Meta’s model edges out Google’s in both intelligence and cost per task at these specific settings.
However, Google holds a significant advantage in throughput. Artificial Analysis measures Gemini 3.8 Flash "high" at approximately 305 output tokens per second, considerably faster than Muse Spark’s 235 tokens per second—a difference of roughly 30%. Furthermore, Gemini currently boasts a lower raw API sticker price. Google is offering an introductory rate of $0.75 per million input tokens and $3.75 per million output tokens. Meta’s corresponding rates are $1.25 and $4.25. It’s important to note that Google’s promotional pricing for Gemini 3.8 Flash expires on December 31, after which it will increase to $1.50 per million input tokens and $7.50 per million output tokens.
This comparison offers a valuable snapshot of the increasingly competitive economics at the frontier of AI model development. Currently, Meta leads this independent comparison by two Intelligence Index points and three cents per benchmark task. Google, on the other hand, provides substantially higher output throughput and more affordable raw token costs during its introductory promotion. Wang’s "Gemini who?" quip might be entertaining executive banter, but for enterprise architects, the reality is more nuanced: Gemini currently presents a faster option, while Muse Spark, by this independent measure, offers a slightly more capable high-effort agent.
Meta’s Evolving Stance on Open Weights
Beyond the immediate benchmark race and pricing strategies, a more significant concern for some developers revolves around Meta’s shifting position on open-weight models. When Meta launched Muse Code and Muse Spark 1.2 in August, there was a notable departure from the open-weight strategy that had propelled Llama to widespread adoption. Muse Code and Spark 1.2 were proprietary, API-served products, a stark contrast to Meta’s previous advocacy for open AI as the path forward.
However, Meta altered its course again just five days later. On August 10, the company released the 30-billion-parameter Muse Glimmer under an Apache 2.0 license, explicitly stating its return to open source. Zuckerberg also announced plans to open the weights for Muse Spark 1.2 in the coming weeks, a plan that was also reported by Reuters.
Now, with the release of Muse Spark 1.3 as another proprietary model, the roadmap for open-weight releases has become less clear. Meta’s latest announcement no longer specifies Muse Spark 1.2 for its open-weight release. Instead, it mentions "the Muse Spark open weights release" without providing details on the version, release date, model size, or license. Zuckerberg’s comments on X also refer to "Muse Spark open weights releases" coming soon.
For development teams that have standardized on models like Llama due to the ability to download weights for self-hosting, customization, and control over inference economics, this ambiguity could be more consequential than marginal gains on composite benchmarks. Muse Spark 1.3 clearly demonstrates Meta’s capacity to iterate proprietary frontier models at an extraordinary pace. The currently available "xhigh" configuration is fast, competitively priced, and significantly closer to the top of independent rankings than its predecessors. The "max" preview indicates Meta’s ability to push the model’s capabilities further when allowed to allocate more reasoning compute.
The critical next test for Meta lies in its ability to translate this impressive iteration speed into a predictable and actionable roadmap for enterprises. This includes making its most advanced capabilities broadly deployable and, crucially, delivering the open-weight Spark model it has already indicated is forthcoming. The ongoing evolution of Meta’s strategy will be closely watched by the AI development community.
