Meta unveiled Muse Spark 1.1 on July 9, a major update to its flagship AI model targeting agentic coding and automation workloads, as the company moves to compete directly with Anthropic and OpenAI in the paid developer-tools market 23.
Alexandr Wang, Meta's chief AI officer, said Muse Spark 1.1 is the company's "strongest model for agentic and coding work yet" 3. The release comes three months after Meta introduced the original Muse Spark model in April, which was available only to select partners through a private API preview 3.
Meta is now making the Muse Spark 1.1 API available through a developer portal as part of a public preview, where users can sign up and see instructions for integration 3. Some early partners can already access the API, and new users can join a waitlist for access over time 3.
Meta will charge $1.25 per million input tokens and $4.25 per million output tokens for Muse Spark 1.1, with $20 in free credits for each new API account 3. Wang characterized the pricing as "very aggressive and attractive" compared with similar offerings from labs like Anthropic and OpenAI 3.
[ANALYSIS] Requiring developers to pay for access to Muse Spark 1.1 represents an effort to diversify Meta's revenue, as the company now asks developers to pay for its new AI model.
Meta describes Muse Spark 1.1 as a "step-change" from the first generation, with improvements based on developer feedback 4. The company says the model is capable of more advanced coding, including detection and fixing of complex bugs 42. Meta's pitch to users centers on Spark's ability to handle large agentic workloads, fix bugs, and help with large code migrations — the kind of automation that enterprises are increasingly turning to AI companies to provide 2.
The model also better supports end-to-end agentic workflows across a range of apps, including multi-agent systems, and has native multimodal perception across images, videos, and documents 4. Wang said Muse Spark 1.1 outperformed rival models in certain tasks involving interaction with third-party coding products and services 3.
Meta is limiting Muse Spark 1.1 API access to its own properties rather than third-party platforms like OpenRouter 2. The company is also developing a variant of Muse Spark that it intends to open source 3.
Separately, Meta released Muse Image, originally code-named Mango, earlier in the week 3. The company is also currently training a more powerful AI model code-named Watermelon 3.
Meta's MSL unit trained Muse Spark 1.1 to excel in coding-related tasks to improve AI agents' overall capabilities 2.
[ANALYSIS] The simultaneous push into paid API access, open-source variants, and next-generation model training signals a multi-track strategy as Meta attempts to establish itself in a coding-AI market where Anthropic and OpenAI hold early positions.