On July 9, 2026 (US time), Meta officially released Muse Spark 1.1, a new AI model built for agent-style coding. It is offered to developers for a fee through the Meta Model API, marking the first time Meta has charged for the use of its own model. With aggressive pricing, the company is setting out to chase down front-runners OpenAI and Anthropic.

A multimodal model built for agentic work

Muse Spark 1.1 is a multimodal model (able to handle several types of data, such as text and images) developed by Meta Superintelligence Labs, and it was first announced back in April. It has now become available in a form developers can actually put to use.

Meta says the model performs strongly on "personal agentic tasks" that involve multi-step reasoning and the planning and orchestration needed to work across external apps and services. The intended use cases are practical ones: large-scale agentic workloads, bug fixing, and sizable code migrations.

On the technical side, Meta highlights a self-managed context window of up to 1 million tokens, and the ability to operate as either a primary agent or a subagent within setups where multiple AIs cooperate. The model also natively supports MCP (Model Context Protocol) servers and custom skills for connecting to external tools, and it can directly control a computer, according to the company.

A "low-price" strategy aimed squarely at rivals

The point Meta keeps stressing is affordability. Using Muse Spark 1.1 through the Meta Model API costs 1.25 USD (about 200 yen) per million input tokens and 4.25 USD (about 690 yen) per million output tokens. That sits just slightly above the rates for Anthropic's Claude Haiku 4.5 and OpenAI's GPT-5.6 Luna.

The Meta Model API opened the same day as a public preview for developers in the US. It uses pay-as-you-go metered billing, and new accounts are reported to receive 20 USD (about 3,200 yen) in free credits. ※1 USD = 162 JPY (as of the July 10, 2026 close)

Until now, Meta has led with a strategy of distributing its AI models for free. Introducing a mechanism that charges for the use of its own model is being read as a clear signal that Meta is making a serious entry into the paid-AI-model market where OpenAI and Anthropic have a head start.

Zuckerberg returns to X after three years to make the pitch himself

To coincide with the release, Meta CEO Mark Zuckerberg posted on X (formerly Twitter) himself. It was his first use of X in about three years, since the service was renamed from Twitter in July 2023. In the post, he described Muse Spark 1.1 as "a strong agentic and coding model at a very low price," playing up its performance on agentic work and tool use.

Zuckerberg also said there was "more to come soon," signaling further model releases. Indeed, Meta unveiled an image-generation model called Muse Image the same week, making its intent to ship models in rapid succession clear. New models from other companies, including OpenAI's GPT-5.6, arrived around the same time, intensifying the competition over developer-facing AI.

Summary

Meta Muse Spark 1.1 is a strategic model that centers on agentic work and coding and chases OpenAI and Anthropic with a new play: low pricing plus a paid API. With features such as a self-managed 1-million-token context window and MCP support, it lines up specs that should resonate with developers building agents in the real world. How far Meta—having switched from free distribution to charging—can win developers over on the strength of price competitiveness looks like the thing to watch from here.