Top AI Product

Meta shipped Muse Spark 1.3 on September 2, and Zuckerberg calls it the company’s biggest jump ever in coding and agentic capability. HackerNews agrees enough to push it to 439 points.

What actually changed

This is a frontier LLM tuned for long-horizon, multi-agent workflows — the kind where an agent grinds through a codebase for hours. Versus Spark 1.2: roughly 20% fewer tool calls and 25% fewer tokens for the same work. That’s not a vanity benchmark; for agents, verbosity is literally your bill. At $1.25 input / $4.25 output per million tokens with a 1M context window, “frontier performance at low cost” is a direct shot at Anthropic’s pricing.

How to use the API

It’s live in Muse Code (Meta’s multi-agent coding terminal), the Meta Model API, and OpenRouter as meta/muse-spark-1.3 — all OpenAI SDK-compatible, so swapping it into an existing agent stack is a one-line change. Typical use case: long-running coding agents where token burn kills the economics.

The real story is the promise of an open-weights release. A frontier-grade agentic model with downloadable weights would reset the whole pricing floor. No date yet — but that’s the thing to watch.

You Might Also Like

• Meta Model api Muse Spark 1 1 Undercuts Openai and Anthropic at 1 25 4 25 per Million Tokens
• Meta Muse Code Muse Spark 1 2 59 on Deepswe and 0 20 Output Tokens if you Share Your Data
• Meta Muse Glimmer 30b Open Agentic Model Apache 2 0 Fits a Full Agent Loop Into one 24gb gpu
• Meta Muse Spark 14b and 9 Months Later Alexandr Wang Delivers but its Closed Source
• Zcode Zhipu z ai glm 5 2 Coding Agent hit hns 1 Spot Chinas Open Answer to Claude Code