Artificial Intelligence

Meta launched Muse Spark 1.3 on September 2, 2026, reporting 25% fewer tokens and 20% fewer tool calls versus its predecessor, though these figures come from internal comparisons with no independent reproduction. Artificial Analysis independently scored the generally available variant at 61 on its Intelligence Index, but did not reproduce the specific DeepSWE v1.1 and MRCR v2 figures cited in Meta's release. The most capable max reasoning mode remains in limited partner preview with no public pricing or confirmed safety-testing completion date.

Anthropic's September 1, 2026 release of Claude Fable 5.1 drops the cache-read price from $1 to $0.25 per million tokens while keeping base input and output rates at $10 and $50, respectively. Anthropic projects overall cost reductions of roughly 25% for typical workloads and up to 45% for highly agentic ones, though the underlying cache-hit-rate assumptions behind those estimates are not publicly disclosed. The release also contains a one-day discrepancy in Anthropic's own documentation over which accounts are subject to the new context-binding enforcement rules.

Broadcom reported $16.7 billion in AI-semiconductor revenue for Q3 FY2026, up 221% year over year, and issued management forecasts of approximately $115 billion for fiscal 2027 and $230 billion for fiscal 2028. These figures are forward-looking outlooks, not disclosed contracted backlog, and CEO Hock Tan identified deployment readiness, advanced wafers, substrates, and high-bandwidth memory as binding constraints even as the company states it has secured supply. A Singapore substrate facility operated by TOPPAN subsidiary AST is targeted to begin production by end of 2026 and is expected to partially address the substrate bottleneck.

Taiwan's Fair Trade Commission is watching TSMC's position in semiconductor markets but has taken a self-described reserved approach, with no investigation opened and no enforcement planned as of September 2, 2026. TSMC holds 73% of the global pure-foundry market by one measure, yet the public record contains no case number, complaint, or formal legal classification tied to this scrutiny. The distinction matters: regulatory awareness and active enforcement carry very different implications for the chip supply chain and the companies that depend on it.

On September 1, 2026, the Department of Justice filed a Statement of Interest in the consolidated OpenAI copyright litigation, arguing that copying works to train large language models is transformative and likely fair use under all four statutory factors. The filing draws a firm line between the training stage and model outputs, leaving questions about data acquisition methods and potentially infringing outputs unresolved. The brief does not bind the court or create a safe harbor, and summary-judgment briefing from the parties does not open until September 24, 2026.