Meta Enters AI Coding Competition Weeks After Anthropic, Trailing OpenAI

Mumbai, July 10: Meta is making its move in the AI coding competition. The tech giant released a substantial update to its model in an attempt to gain ground on competitors OpenAI and Anthropic. The latest product unveiling comes three months after Meta introduced its first AI model when it announced plans to the public. 

Developed by Meta’s head of AI, Alexandr Wang and his team, Muse Spark 1.1, which was released Thursday, Meta’s “strongest model for agentic and coding work yet,” according to Wang, a CNBC report said. The previous iteration of Muse Spark launched last month in April was available only to select partners via private API preview.  

This release, however, Meta is making its model’s API available through a developer portal as part of a public preview so users can register and follow set instructions to integrate the API. “Some of our earliest partners have access today. Additional users will be graduated from our waitlist over time,” Meta spokesperson said. Currently, partners can access Muse Spark only on Meta’s properties, not third-party platforms such as OpenRouter, a popular marketplace. 

“This is going to be served on top of the computer infrastructure that we’ve built,” Wang said. 

Also Read: How to invest in US stocks  

It’s Meta’s second major AI launch this week, following Tuesday’s launch of Muse Image, which aims to target creatives and advertisers with its picture-generation capabilities. The speed of developments come as Zuckerberg faces increased pressure to demonstrate a return on Meta’s enormous investments into AI compute infrastructure.  

While Meta has matched hyperscaler peers in raw infrastructure spending, it doesn’t have its own cloud business — one is reportedly being built — and has fallen behind OpenAI, Anthropic and Google in both creating buzzworthy models and commercial applications. 

“The pricing will be very aggressive and attractive compared to what’s out there,” Wang told CNBC of Muse Sparks. All API accounts come with $20 worth of free credits. After that, Meta will charge $1.25 per million input tokens and $4.25 per million output tokens. “The goal is to really have attractive pricing that makes sense for people that are consuming at an immense level,” Wang said. He said Muse Spark 1.1 is faster than competitor models on certain benchmarks when called upon to integrate with outside coding tools. 

Also Read: US Stock market timings  

Muse Spark was trained to be good at coding because building that competency is foundational to good agentic performance overall. With great coding capabilities, AI can perform tasks like a “serverless labor force” of digital interns. Meta designed Muse Spark 1.1 to integrate easily with the most widely-used developer “harnesses” today including those built on OpenRouter and OpenAPI, technologies that have surged in popularity after OpenClaw gained popularity this year, CNBC reported. 

Meta is also marking a departure from its previous practice of open-sourcing its Llama AI models first and then selling access to behind-firewall systems. “We are still committed to open source. There’s a version of Muse Spark that we are building that we will open source,” Wang said, though he declined to comment on when that model would be made public. Wang said he’s been testing out Muse Spark 1.1 himself for health-related use cases including searching the internet, scanning academic papers and pulling out his own health information. Wang confirmed that Meta is training a next-generation model dubbed Watermelon, though a target release date has not been shared.