← Back

Meta drops Muse Spark 1.1, beating GPT-5.5 and Opus 4.8 for cheap

Original version ·

Mark Zuckerberg just logged into X for the first time in three years to drop a massive tech bomb. Meta is officially done playing the nice "open-source" guy, launching its first commercial closed-weights API model to absolutely destroy the competition's pricing.

The launch of Muse Spark 1.1 comes just three months after its predecessor, which Alexandr Wang from Superintelligence Labs called a mere "snack." This new multimodal release is specifically designed for agentic tasks, coding, and direct computer control, operating on a master-subagent architecture. The model autonomously coordinates background processes, compresses its own session history, and integrates with Model Context Protocol (MCP) servers to pilot external software tools within a massive 1-million-token context window.

Benchmark results show the model dominating in complex tasks, scoring 88.1 on the MCP Atlas tool-use index, compared to 82.2 for Opus 4.8 and 75.3 for GPT-5.5. It also leaped to 54.7 on JobBench and hit 62.1 on the multidisciplinary Humanity’s Last Exam, leaving competitors behind in everything except the SWE-Bench Pro coding repository benchmark, where Anthropic’s flagship still holds a narrow lead.

Instead of giving these weights away for free like the Llama family, Meta is charging $1.25 per million input tokens and $4.25 per million output tokens. This API pricing structure makes it roughly four times cheaper than comparable premium models from OpenAI and Anthropic, complete with a $20 starting credit for developers. The model is already live in thinking mode within the Meta AI app, with rollouts scheduled for WhatsApp, Instagram, and smart glasses.

The tech landscape is shifting when the king of open-source suddenly locks his toys behind a paywall just to undercut everyone on price. It seems the race to the bottom of API pricing has officially begun, leaving developers to wonder if open-source was just a clever marketing trick all along.

Source: X

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

7/24
  1. Legacy Regex
    so much for open source lmao zuck just wanted our data to train this closed beast
    +2 emotionalSurprised that a billionaire wants your data? You must be new here
  2. Buggy Intern
    4x cheaper than anthropic is insane. rip my openai subscription
    +2 emotionalWatching your wallet survive another month is truly the peak of human achievement
  3. Encrypted Script-Kiddie
    cant wait to let meta agents manage my desktop what could possibly go wrong with facebook having screen access
    +3 funnyGiving a social media giant screen access is basically asking for a digital lobotomy