Strategy
69 analyses
The first model in the Mythos class available to the public arrives at the API for $10 per million input tokens and $50 for output, achieving 80.3% on SWE-bench Pro compared to 58.6% for GPT-5.5.
Jensen Huang was in Seoul to announce a package ranging from HBM memory to humanoid robots, with a gigawatt data center from SK Telecom slated for 2027 as the most visible piece.
At WWDC 2026, Craig Federighi introduced a rebuilt Siri based on a customized Gemini with 1.2 trillion parameters and opened Apple Intelligence to Anthropic, OpenAI, and xAI.
Microsoft AI CEO stated to Bloomberg that Claude is too expensive and the goal is to eliminate the cost. Build 2026 introduced seven proprietary models without distillation from OpenAI or Anthropic.
In his last conference as CEO, Tim Cook announced an Extensions system in iOS 27 that allows users to choose Claude, Gemini, or ChatGPT as the engine for Apple Intelligence. A $1 billion per year agreement with Google becomes the default.
The Japanese government confirmed on June 3 that the three megabanks have access to Anthropic's Claude Mythos. Mizuho is already planning to cut 5,000 back-office positions over ten years with an investment of $320-640 million in AI.
Financial Times reported on June 5 that around six engineers from Anthropic were deployed to the NSA to adapt the Mythos model for offensive operations while the company litigates with the Pentagon.
The announcement on 3 June crystallises the sector's paradox: Indian services purchase productivity from AI for the offshore base that sustains margins, while the European arm trims the middle of the pyramid.
The open model of 550 billion parameters arrived at Hugging Face on 4 June. SAP, ServiceNow, Salesforce, Adobe, and CrowdStrike run agents within NVIDIA's OpenShell sandbox.
AI custom chip revenue grew by 143% year-on-year in the second fiscal quarter. Anthropic plans to purchase 1 GW in TPUs in 2026 and 3 GW in 2027, according to CEO Hock Tan, indicating a direct signal regarding the laboratory's IPO.
MAI-Thinking-1 features 35 billion active parameters, a context window of 256,000 tokens, and was trained without the distillation of third-party models, in a move Mustafa Suleyman described as true self-sufficiency.
Office 365 Copilot gains Agent Mode as the default state in the June rollout. Azure AI Foundry transforms into a control tower for orchestration at scale, and Windows Local AI redefines data sovereignty friction.