Publishers file class-action suit over Llama training data
Five major publishers—Elsevier, Cengage Learning, Hachette Book Group, Macmillan Publishers, and McGraw Hill—alongside author Scott Turow and his company S.C.R.I.B.E. Inc., filed a proposed class-action complaint against Meta Platforms and chief executive Mark Zuckerberg in the US District Court for the Southern District of New York on May 5. The complaint alleges Meta copied millions of copyrighted books and academic articles without permission to train its Llama artificial-intelligence models.
The plaintiffs allege Meta obtained the material from online pirate libraries and other unauthorised sources, including through torrent downloads, and removed or disregarded copyright-management information. They are seeking damages and an injunction, including destruction of unauthorised copies retained by Meta. Meta has said it will contest the lawsuit vigorously, arguing that training AI systems on copyrighted material can be protected under the US fair-use doctrine. Zuckerberg is named personally, with the plaintiffs alleging he knew about and authorised decisions concerning the acquisition and use of copyrighted training material. The case remains a proposed class action and has not been certified by the court.
Muse Spark 1.1 launch and the move to a paid API
Meta Superintelligence Labs released Muse Spark 1.1 on July 9, 2026, alongside a public preview of the Meta Model API. The model is multimodal and built for coding and agentic work, and marks the first time Meta has sold model access through a metered, proprietary API rather than distributing open weights. Meta Chief AI Officer Alexandr Wang called it the company's strongest model yet for agentic and coding work, in comments captured on CNBC's launch coverage. The release breaks from the strategy that made Llama the default base for startups and research teams that could not afford OpenAI or Anthropic API bills.
Pricing is set at $1.25 per million input tokens and $4.25 per million output tokens, with every new developer account receiving $20 in free credits before billing switches to pay-as-you-go, according to Meta's announcement and CNBC's reporting. Meta has not published open weights for the Muse Spark family, and coverage of the launch framed the move explicitly as a shift to a paid API business rather than another open-weight release.
Benchmark position and competitive pricing
On the Artificial Analysis Intelligence Index, Muse Spark 1.1 scored 80.0, placing it behind Anthropic's Claude Opus 4.8 at 82.7 and OpenAI's GPT-5.5 at 83.4, according to DataCamp's July 9, 2026 launch report. Despite the benchmark gap, a Yahoo Finance analysis found Muse Spark 1.1 runs roughly 75% cheaper than Claude Opus 4.8 on input tokens and 83% cheaper on output tokens. Against GPT-5.5, the gap is wider, at roughly one-tenth the cost. TechCrunch reported that the pricing actually sits closer to Anthropic's budget-tier Claude Haiku 4.5 and OpenAI's GPT-5.6 Luna than to the frontier tier, suggesting Meta is competing on cost against the cheap, high-volume tier every major lab now offers while claiming performance closer to the expensive tier.
Industry backdrop: surging training-data volumes
The publisher lawsuit lands against a backdrop of rapidly expanding training-data volumes across the industry. Meta reported Llama 3.1 was trained on more than 15 trillion tokens, while model documentation for Llama 4, released in April 2025, stated its Scout and Maverick models were pre-trained on as many as 40 trillion tokens. Chinese developer DeepSeek used 8.1 trillion tokens to train DeepSeek-V2 and 14.8 trillion for DeepSeek-V3, per Epoch AI's database. Token totals do not establish whether underlying material was lawfully acquired, but they illustrate the scale of datasets now required for frontier systems. A June 2025 ruling by US District Judge Vince Chhabria in Meta's favour on fair-use claims brought by authors including Sarah Silverman has distinguished training from data acquisition, a distinction the new publishers' complaint seeks to exploit by alleging pirate-source acquisition.
Broader AI monetisation push across Meta's apps
Separately, Meta is preparing to trial premium subscriptions across Instagram, Facebook, and WhatsApp that would unlock expanded AI capabilities and additional features, according to TechCrunch reporting from January 27, 2026. The subscriptions are separate from Meta Verified and would include the AI-generated short-form video experience Vibes—free since its September 2025 launch—moving to a freemium model that locks some video creation behind a paid tier. The Manus suite of general AI agents, acquired in December for a reported $2 billion, would also be folded into the subscription plans, with Meta integrating Manus into its products while continuing to offer it as a standalone subscription to businesses. Pricing for the new tiers was not disclosed. The challenge, as one analysis noted, will be persuading users that AI features on social media warrant paying for them, given that the Llama family has historically remained open-source and free.
Share this article
