Anthropic weekly, July 20-27: Opus 5, AMD's $5B bet, and the copyright settlement

Anthropic used July 20-27 to push Claude up the capability ladder and deeper into institutions: Opus 5 launched at Opus 4.8 pricing, voice mode reached Opus and Sonnet, AMD committed up to $5 billion alongside a 2GW GPU plan, Anthropic announced $200 million for economic-transition research, and a federal judge approved the company's $1.5 billion authors' settlement.

Claude Opus 5 launched on July 24 across Anthropic's platforms. Anthropic says it is the new default for Max subscribers and the strongest model on Pro, with performance near Fable 5 on many coding and knowledge-work evaluations. The company kept the Opus 4.8 price: $5 per million input tokens and $25 per million output tokens. Fast mode runs at about 2.5 times the default speed for twice the base price. 1

The commercial pitch is efficiency rather than a new price tier. Anthropic says Opus 5 more than doubles Opus 4.8's Frontier-Bench performance at a lower cost per task and approaches Fable 5 on CursorBench at half the cost per task. Those are Anthropic's own evaluations, so buyers should treat them as vendor-reported benchmarks. The safety boundary is also explicit: Opus 5 remains behind Mythos 5 on biology and offensive cybersecurity, and some flagged cyber requests fall back to Opus 4.8.

Anthropic also expanded voice mode beyond Haiku. Opus and Sonnet can now handle voice conversations, switch between voice and text mid-conversation, and connect to workflows involving Gmail, Slack, and Canva. The rollout adds French, German, Spanish, Hindi, Indonesian, Italian, Japanese, Korean, and Portuguese beyond the previous English-first experience. The change is an expansion of a voice product that launched earlier, not a new voice-mode debut. 2

On July 22, Anthropic launched an Economic Index connector for claude.ai. Users can enable it from the connector directory and ask questions about how people use Claude by occupation, location, task, or time period. It works with any Claude model and returns answers grounded in the Index data, with the underlying datasets still available on Anthropic's site. Anthropic cautions that the Index measures Claude usage, not the labor market as a whole. 3

AMD said it will invest up to $5 billion in Anthropic as part of a multi-year infrastructure partnership. Anthropic plans to deploy up to 2 gigawatts of AMD Instinct MI450 GPUs through AMD's Helios rack-scale system, with the first gigawatt targeted for the first half of 2027. AMD will also use Claude across software, engineering, and product development. The agreement is a planned strategic investment and compute deployment, not a new Anthropic equity round disclosed on the terms of the May Series H financing. 4

The Economic Futures Research Fund will support external research on how to prepare for AI-driven disruption. Anthropic says it will prioritize workplace integration, worker transitions, income support, worker stakes in AI-driven growth, and public investment. It expects most grants to fall between $5 million and $30 million and says the fund will not directly fund proposals below $1 million. The program is global, but applicants must be accredited institutions, independent research institutes, policy organizations, or nonprofits with a record of large-scale field experiments. 5

The company also announced another $20 million donation to Public First Action, bringing its total support for the nonpartisan AI policy and public-education organization to $40 million. Anthropic says the donations cannot be used to influence the election of a federal, state, or local candidate. This is policy spending, not research funding or venture financing. 6

A smaller program targets rare genetic disease research. Anthropic will offer accepted scientists and early-stage biotechs up to $50,000 in Claude credits over six months, with applications due August 2, 2026 at 11:59 PM Pacific time. One track focuses on basic science and works with the Monarch Initiative; the other supports biotech teams working on clinical development. 7

On July 20, U.S. District Judge Araceli Martinez-Olguin in San Francisco granted final approval to Anthropic's $1.5 billion class-action settlement with authors who alleged that the company used pirated books to train Claude. The settlement resolves claims over the storage and use of more than 7 million pirated books; a prior ruling held that training on the books was fair use but found that maintaining the pirated central library violated authors' rights. Reuters reported that more than 91% of covered authors and publishers had claimed their payments. Some authors and publishers opted out and continue to pursue separate cases. 8

No new equity financing round, executive leadership change, or newly named enterprise customer announcement was verified for this seven-day window. The week's capital story was AMD's planned infrastructure investment, while the customer-facing product changes were Opus 5, voice mode, and the Economic Index connector.

The immediate dates are concrete. Rare-disease grant applications close August 2. AMD's first gigawatt target is the first half of 2027. The legal settlement moves into distribution while opt-out cases continue. For buyers, the practical question is whether Opus 5's lower task cost and broader voice access translate into production adoption without changing the safety and fallback controls that remain around high-risk cyber and biology work.