Meta released Muse Spark 1.1 on July 10, 2026, the latest AI model from its Meta Superintelligence Labs (MSL) division, marking the company's effort to establish itself as a competitive force in the frontier AI market. The release aims to position Meta against rivals through advanced agentic and coding capabilities combined with aggressive pricing. The launch coincides with a broader surge of AI announcements this week, including new model families from OpenAI and xAI, highlighting the accelerating pace of competition across the industry.
Meta Releases Multimodal AI Model With One-Million-Token Context Window
Muse Spark 1.1 is positioned as a multimodal reasoning model optimized for agentic tasks — those requiring sustained planning, tool use, and multi-step execution across external applications and services. The model supports a one-million-token context window and is trained to manage that context actively, compacting information and retrieving relevant details across extended sessions without losing coherence. According to Meta, it generalizes in a zero-shot manner to new native tools, MCP servers, and custom skills, and can operate both as a primary orchestrating agent and as a delegated subagent within larger systems.
In terms of computer use, Muse Spark 1.1 is designed to navigate multi-application workflows where information changes dynamically. Rather than executing every action through the interface, it selects between writing automation scripts and direct interaction depending on what is more efficient — a behavior Meta says was deliberately trained into the model. On the coding side, the update brings substantial gains on enterprise-scale tasks: diagnosing complex bugs, implementing features in large codebases, and executing code migrations.
MSL chief Alexandr Wang noted in media reports that coding capability is treated as foundational to agentic performance rather than a standalone feature. "You kind of have to build coding capabilities as part of that in service of overall agentic capabilities," he said.
The model also advances multimodal understanding, with strengths in visual-to-code generation, image and video captioning, and agentic workflows that combine perception and action. Developers using early API access have described it as a complete agentic foundation capable of handling large-scale workloads — a characterization that aligns with Meta's stated ambition of building toward what it calls "personal superintelligence."
Meta opened a public preview of the Meta Model API alongside the release, allowing developers to begin building with the model directly.
Meta Undercuts Competitors With $1.25 Input Token Pricing
Meta is entering the API market at $1.25 per million input tokens and $4.25 per million output tokens — figures that Wang characterized as "very aggressive and attractive" relative to competing frontier models. New accounts will also receive $20 in free credits. By comparison, leading models from Anthropic and OpenAI are typically priced two to five times higher on output tokens, placing Muse Spark 1.1 in a substantially different cost category for high-volume use cases.
This pricing strategy signals something broader than a product launch. Meta is making an explicit bid to attract enterprise developers and high-consumption users who have until now been constrained by the operational cost of frontier-model inference. For organizations running large agentic workloads — the kind that require sustained multi-step reasoning, continuous tool calls, and long context retention — output cost is often the dominant variable in total expenditure. A model that performs competitively at a fraction of the price is not simply a cheaper alternative; it changes the economic calculus of what can be built and at what scale.
Whether this constitutes the opening of a sustained price war remains to be seen, but the pressure on competitors is real. Anthropic, OpenAI, and Google have all made recent investments in lower-cost model tiers, and the trajectory of the market has been consistently toward declining inference costs. Meta's entry at this price point may accelerate that trend. Wang indicated that the goal is to "have attractive pricing that scales with immense consumption usage" — a framing that suggests Meta is optimizing for volume adoption rather than margin, a posture its hyperscaler competitors will need to respond to.
What is clear is that the frontier AI market is becoming difficult to navigate on capability alone. As models converge in benchmark performance, pricing, developer experience, and ecosystem integrations are emerging as the decisive differentiators — and Meta, with its infrastructure scale and appetite for aggressive investment, is now a serious participant in all three.
FAQ
What did Meta release on July 10, 2026?
Meta released Muse Spark 1.1, the latest AI model from its Meta Superintelligence Labs (MSL) division. The model is optimized for agentic and coding applications and supports a one-million-token context window. Meta also opened a public preview of the Meta Model API alongside the release.
How much does Meta charge for Muse Spark 1.1 API access?
Meta is entering the API market at $1.25 per million input tokens and $4.25 per million output tokens. New accounts will receive $20 in free credits. According to MSL chief Alexandr Wang, this pricing is "very aggressive and attractive" compared to competing frontier models from Anthropic and OpenAI, which are typically priced two to five times higher on output tokens.
What capabilities does Muse Spark 1.1 offer for developers?
Muse Spark 1.1 is designed for agentic tasks requiring sustained planning, tool use, and multi-step execution across external applications. It supports multimodal understanding including visual-to-code generation, image and video captioning, and can navigate multi-application workflows. The model brings substantial gains on enterprise-scale coding tasks such as diagnosing complex bugs, implementing features in large codebases, and executing code migrations.
