Jul 10, 2026 10:52:00

Meta announces its proprietary AI 'Muse Spark 1.1,' achieving benchmark scores equivalent to Claude Opus 4.8.

Meta's AI research arm, Meta Superintelligence Labs, has announced ' Muse Spark 1.1, ' a multimodal inference model. Positioned as a major upgrade to '

Muse Spark ,' which was released in April 2026, it represents significant progress in tools, PC operation, coding, and multimodal understanding.

Introducing Muse Spark 1.1
https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/

Muse Spark 1.1 is a high-performance model for personal tasks that run across external apps and services. It can actively manage a 1 million token context window, remembering past operations throughout long sessions, retrieving information from much earlier work, and processing tasks while maintaining the steps necessary for subsequent tasks.

Coding performance has been significantly improved in real-world tasks dealing with large and complex codebases, and it shows performance far superior to the original model in applications such as creating web applications and end-to-end question answering.

In Meta's proprietary coding evaluation tool, Meta Internal Coding Bench, Muse Spark 1.1 significantly outperformed Muse Spark, surpassed GPT-5.5, and demonstrated performance comparable to Claude Opus 4.8.

The benchmark test results for 'Muse Spark 1.1,' 'Muse Spark (previous generation model),' 'Gemini 3.1 Pro,' 'Claude Opus 4.8,' and 'GPT-5.5' are shown below. The Muse Spark 1.1 recorded the top score, surpassing the Claude Opus 4.8 in multiple tests.

Muse Spark 1.1 is being used daily by developers and researchers throughout Meta.

In addition to coding and agent functionality, Muse Spark 1.1 excels in perception, multimodal reasoning, and tool utilization. Below is an example of how it extracts photos from a video taken on a smartphone, infers information about a product, interacts with the user's browser, and lists products on Facebook Marketplace on the user's behalf.

Muse Spark 1.1 also excels in perception and multimodal reasoning, inspecting visual and audio inputs, preserving details across long workflows, and acting on them in real execution environments. It shows particular strengths in visual-to-code generation, rich image/video… pic.twitter.com/GkebF7gcEd

— AI at Meta (@AIatMeta) July 9, 2026

Muse Spark 1.1 is available in the Meta AI app and meta.ai's 'Thinking' mode, and can also be used for development via the Meta Model API.