26 April 2026: Google Bets $40 Billion on Anthropic as DeepSeek V4 Goes Open Source (AM)

This weekend produced some of the biggest financial commitments and model launches in recent AI history. Google has agreed to invest up to $40 billion in Anthropic, China’s DeepSeek has dropped a new open-source flagship that rivals the frontier, and Canada and Germany have forged an alliance to build AI that sits outside Silicon Valley’s control. Here is what you need to know heading into Sunday.

Google has committed up to $40 billion to Anthropic, making it the largest single investment in an AI company to date. Alphabet will put $10 billion in immediately at a $350 billion valuation for Anthropic, with a further $30 billion conditional on the Claude maker hitting performance targets. As part of the deal, Google Cloud will supply Anthropic with five gigawatts of computing capacity over the next five years. Amazon has also committed up to $25 billion separately, meaning Anthropic now has the prospect of more than $65 billion in outside investment.

The announcement coincided with reports of Anthropic’s Mythos model, described as its most powerful model to date and built specifically with cybersecurity applications in mind. Anthropic has kept Mythos to a small group of trusted partners, citing the risk of misuse, though the model has reportedly already appeared in unauthorised hands. For UK organisations in finance, defence, or critical national infrastructure, Anthropic’s security-focused direction will be worth watching closely as access is gradually broadened.

DeepSeek released two new open-source models on Thursday that challenge the frontier at a fraction of the cost. DeepSeek V4-Pro packs 1.6 trillion total parameters but activates only 49 billion at a time — a design called a Mixture of Experts that keeps it fast and efficient to run. V4-Flash is the leaner sibling at 284 billion parameters total. Both support a one-million-token context window, meaning they can process roughly the equivalent of 750,000 words in a single session. Both are available as open weights under the MIT licence and via the DeepSeek API from day one.

For small businesses and independent developers, the pricing is the headline. Flash costs $0.14 per million input tokens; Pro costs $1.74 — well below comparable tiers from OpenAI and Anthropic. Independent benchmarks suggest V4-Pro falls only marginally short of GPT-5.4 and Gemini 3.1-Pro on coding and reasoning tasks. A one-million-token window means you could feed an entire document library, codebase, or contracts archive into a single API call. For anyone monitoring their AI costs, V4-Flash is worth a serious look.

Canadian AI company Cohere has agreed to merge with Germany’s Aleph Alpha in a deal valuing the combined group at around $20 billion. Aleph Alpha, which has built AI systems for European governments and regulated industries, will join forces with Cohere’s enterprise platform. The Schwarz Group — the conglomerate behind Lidl and Kaufland — is contributing $600 million and will run the new entity on its own sovereign cloud infrastructure. Dual headquarters will operate in Toronto and Heidelberg.

The strategic intent is clear: both companies believe there is significant demand for AI that is not built, hosted, or controlled by US technology companies. European governments have been cautious about embedding American-built models in sensitive public services, and the merged entity plans to target defence, finance, healthcare, and telecoms specifically. For UK organisations that have been reluctant to lock into US providers, this deal creates a credible transatlantic alternative with government backing on both sides. It also follows a joint AI cooperation declaration signed by Canada and Germany in February.

Anthropic ran a closed experiment last week in which AI agents acted as both buyers and sellers inside a real marketplace, completing genuine transactions with real money. The project, described internally as a “classified marketplace” rather than a simulation, was designed to test how agents negotiate, evaluate offers, and manage trust without a human approving each step. Agents representing buyers and sellers were given goals and budgets, then left to close deals independently.

Anthropic has not published full results, but the decision to go public with the experiment signals confidence that agentic commerce is closer to practical deployment than most businesses assume. The architecture that let agents buy and sell goods could equally power procurement automation, supplier negotiation, or inventory management. The unresolved legal question of which party is liable when an AI agent commits to a contract on a company’s behalf will need addressing before this capability reaches mainstream use.

Here is everything else worth knowing from this weekend’s AI news.

• OpenAI CEO Sam Altman apologises over Tumbler Ridge — Altman wrote publicly that he is “deeply sorry” after it emerged OpenAI had information about a mass shooting suspect in Tumbler Ridge, Canada, but did not alert law enforcement. [25 Apr]
• Maine’s data centre moratorium vetoed — The state governor rejected a bill that would have imposed America’s first statewide pause on new data centre construction until November 2027, clearing the way for continued AI infrastructure investment. [25 Apr]
• Apple’s incoming CEO is a hardware engineer — John Ternus, Apple’s new chief executive, comes from an engineering background, prompting analysts to suggest the company may refocus strategy on devices after a period of heavy AI services investment. [25 Apr]
• Amateur solves 60-year-old maths problem using ChatGPT — A non-professional mathematician used ChatGPT to crack an open problem from the Erdos collection of unsolved combinatorics challenges. The solution has since been independently verified. [25 Apr]
• ComfyUI raises $30M at $500M valuation — The open-source tool for AI image and video generation has closed a new funding round as creator demand for customisable, locally runnable AI media tools continues to grow. [24 Apr]
• Meta and Thinking Machines Lab talent moves both ways — Reports describe an active two-way flow of staff between Meta’s AI division and Thinking Machines Lab as both organisations build large model teams aggressively. [24 Apr]

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s daily guide to developments in artificial intelligence, published every morning.