Meta has introduced Muse Spark 1.1, a multimodal reasoning model built for agentic tasks, and opened a public preview of its first paid developer API. The release, announced on July 9, 2026 by Meta Superintelligence Labs, arrives with pricing aggressive enough to trigger what analysts are already calling a price war in the red-hot market for AI coding assistants.
The launch marks a sharp strategic shift for the company. Until now, Meta has given away its most powerful models for free. With Muse Spark 1.1, Chief Executive Mark Zuckerberg is betting that undercutting rivals on cost can buy Meta a foothold in the paid developer market dominated by Anthropic and OpenAI. Zuckerberg pledged "aggressive and attractive" pricing for the model, according to Bloomberg, and the numbers appear to back that up. For more context on this story, see our ongoing AI industry coverage.
What Muse Spark 1.1 Actually Does
According to Meta's official announcement, Muse Spark 1.1 is a multimodal reasoning model engineered for agentic work, with what the company described as major gains in tool use, computer use, coding, and multimodal understanding. It is available immediately in "Thinking" mode inside the Meta AI app and on meta.ai, and developers can now access it through a public preview of the new Meta Model API.
A standout feature is the model's ability to manage a context window of 1 million tokens on its own. Meta said Muse Spark 1.1 remembers earlier actions, retrieves information from deep in a session, and compacts its context to preserve the steps that matter for later work. That is a meaningful capability for long-running coding and automation tasks, where losing the thread of a project has long been a weakness of AI assistants.
The model is also trained to orchestrate multi-agent systems. As a lead agent, it gathers context, forms a plan, and delegates execution to parallel subagents, optimizing for end-to-end latency. As a subagent, it sticks to its assigned task, understands the tools available, and knows when to escalate back to the main agent. On Meta's internal coding benchmark, the company said Muse Spark 1.1 is competitive with leading alternatives.
A New Frontier in Computer Use
Muse Spark 1.1 also pushes into computer-use workflows, where an AI operates software interfaces much like a human would. Rather than clicking through every desktop action one step at a time, the model decides when to write an automation script and when direct interaction is simpler, generating batches of actions at each step. Meta demonstrated the model building a web chat app, taking automated screenshots to spot visible failures, and tracing those problems back through the code to fix them.
The release comes the same week Meta launched Muse Image, its first in-house AI image generator, which is now available inside Instagram and WhatsApp. Together, the two products signal an accelerated push by Meta Superintelligence Labs under the leadership of Alexandr Wang, who Fortune reported is central to the company's effort to close the gap with frontier-model leaders.
The Pricing Squeeze
The most consequential detail may be the price tag. According to Meta's developer pricing documentation, Muse Spark 1.1 costs $1.25 per million input tokens and $4.50 per million output tokens, with cached input priced at just $0.15 per million tokens. Those figures place it below xAI's Grok 4.5, which charges roughly $2 per million input tokens and $6 per million output tokens, and well below Anthropic's premium Opus-tier models.
Industry watchers were quick to frame the move as a deliberate squeeze. Business Insider reported that Meta's new model could spark a "massive price war" in the coding market, while The Decoder noted the API pricing is designed to put pressure on both OpenAI and Anthropic. On developer forums, the reception echoed that view, with one widely shared assessment describing the model as offering "Opus-level intelligence for Haiku prices" — a reference to Anthropic's fastest, cheapest model tier.
The competitive pressure is not coming from Meta alone. The aggressive pricing follows the arrival of strong, low-cost models from Chinese developers, with Zhipu's open-source GLM 5.2 drawing particular attention for matching premium coding performance at a fraction of the cost. The Decoder reported that data platform Databricks recently benchmarked coding agents on its own multi-million-line codebase and found GLM 5.2 matched Anthropic's Opus 4.8 at $1.28 per task versus $1.94.
Why It Matters
For Meta, the stakes are clear. The company has trailed Anthropic and OpenAI in the developer market, where programmers have flocked to tools that can write, debug, and refactor code across large enterprise systems. By offering a capable model at a fraction of the cost, Meta is betting that price-sensitive application developers will give its API a serious look.
For the broader industry, the launch signals that the cost of frontier-grade intelligence is falling fast. With Meta, xAI, Google, OpenAI, and Anthropic all now competing on both capability and price — and low-cost open-source alternatives nipping at their heels — the margins on AI inference are compressing rapidly. That is good news for developers building products on top of these models, but it raises hard questions about which companies can sustain the heavy compute costs of training and serving frontier AI over the long run.
Muse Spark 1.1 is available now to developers through the Meta Model API public preview. Meta has not said when the model will move beyond preview, nor whether the aggressive introductory pricing will hold.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
