AI Product Updates: May 26, 2026

AI Product Updates: May 26, 2026

Google I/O aftershocks hit the ecosystem: Gemini 3.5 Flash, XR glasses, and AI-powered Siri. DeepSeek made its 75% price cut permanent. OpenRouter raised $113M at a $1.3B valuation. Claude Opus 4.7 is now generally available, and a code leak points to Claude Mythos launching sooner than expected.

Daily AI Product Updates
2026/5/26 · 23:34
1 订阅 · 1 内容
A busy Tuesday morning in AI. Google's I/O aftershocks are still rippling across the ecosystem, DeepSeek made its price war permanent, OpenRouter hit unicorn status, and Anthropic had one of the noisiest weeks it's seen in months — from a model leak to a cross-platform memory layer to a quietly-tested pricing change that got walked back in hours.
Here's everything that matters for May 26, 2026.

Model releases and updates

Claude Opus 4.7 is now generally available. Anthropic's current flagship ships with a 1-million-token context window and increased image resolution support, accessible across the API, AWS Bedrock, Google Vertex AI, Databricks, and the Claude apps. 1 Pricing is unchanged from Opus 4.6.
In parallel, code leaked to public GitHub repositories on May 22–24 suggests Claude Mythos 1 is closer to launch than Anthropic has publicly indicated. The leaked artifacts include a reasoning protocol nicknamed "Strict Write Discipline," Mythos Router schemas, and references to a production Opus 4.7 thinking variant. 2 One community fork of the leaked material pulled 170 stars in 48 hours. This is the third Mythos-related code leak since March 2026; Anthropic has not confirmed a launch date.
Google's model lineup at I/O 2026 came in three tiers. Gemini 3.5 Flash is the new speed-optimized flagship — tuned for real-time context understanding, faster reasoning, and improved coding support, with deployment across Search, Android, Workspace, and developer tooling. 3 Gemini Omni extends the multimodal range, generating video from text, images from voice prompts, and mixed-media presentations. Gemini Spark is Google's agentic platform layer — designed to handle multi-step tasks like email triage, flight monitoring, and meeting scheduling in the background without continuous user input.
A notably small model also landed this week: Cactus Needle, a 26-million-parameter Simple Attention Network distilled from Gemini 3.1. It runs locally on consumer hardware, hitting 6,000 tokens/sec prefill and 1,200 t/s decode on a Mac. 2 The weights are fully open-source.

Product launches and feature updates

Google I/O 2026 highlights including Gemini 3.5 Flash, Android XR glasses, and AI Search
Google I/O 2026: the full Gemini ecosystem rollout 3
Claude Code shipped to the web and iPhone this week, moving beyond terminal access for the first time. Developers can now run agentic coding sessions from any browser without keeping a laptop open — a direct functional overlap with OpenAI Codex's newly announced ability to control a locked Mac. 2
Claude Memory introduced cross-platform sync: context started in Claude can now be picked up mid-thread in ChatGPT or Gemini, with personalization and preferences carried over. The feature inverts the closed-ecosystem assumption that has structured consumer AI assistants since 2022.
Google's AI-powered Search overhaul — described by Google as the largest change to the product in 25 years — replaces standard results with conversational responses, AI-generated summaries, and intelligent follow-up suggestions. The redesign also debuted Universal Cart, an AI-powered multi-merchant shopping interface that lets users add items across retailers in a single session. 3
Google and Samsung's Android XR glasses were demoed at I/O, showing real-time bilingual translation overlaid directly in the user's field of view. Google Cloud CEO Thomas Kurian separately confirmed that Gemini will power an updated Siri rolling out in iOS 26.4 — Phase 1 in spring, Phase 2 alongside iPhone 18 in September. Apple is paying approximately $1 billion annually to license a custom 1.2-trillion-parameter Gemini model. 2 OpenAI's existing ChatGPT-in-Siri integration moves to secondary.
Alibaba Cloud and Picsart launched an AI Video Competition targeting creative teams, pairing Alibaba's agentic AI ecosystem with Picsart's 130M+ monthly creator base. 4
Anthropic also quietly tested restricting Claude Code to the $100/month Max plan, removing it from the $20/month Pro tier. The change reverted within hours after backlash; an Anthropic employee described it as a limited test affecting roughly 2% of new signups and the public page update as "a mistake." 2

Funding and M&A

OpenRouter raised $113 million in a round led by CapitalG, the investment arm of Alphabet. The deal values the company at approximately $1.3 billion — more than double its valuation from its previous raise. OpenRouter operates as a marketplace providing access to over 400 AI models (including OpenAI, Anthropic, and Google) through a single API, positioning itself as what its founders describe as an AI equivalent of Stripe. 5 The company currently processes 25 trillion tokens per week, up from 5 trillion six months ago.
正在加载内容卡片…

Pricing

DeepSeek made its 75% price cut on V4-Pro permanent. The promotion, originally set to expire May 31, is now the standard pricing. 6 At current rates, V4-Pro frontier reasoning costs roughly one-quarter of equivalent OpenAI or Anthropic tokens. Chinese models already represent 60% of OpenRouter usage; DeepSeek is the largest single contributor. 2
正在加载内容卡片…

Infrastructure and chips

Microsoft raised its 2026 capital expenditure forecast to $190 billion, with $25 billion of the increase attributed to rising memory and storage component prices driven by AI infrastructure demand. Microsoft has spent $97 billion across the past four quarters and generated $37 billion in AI ARR. 2
Amazon disclosed that its custom silicon business — Graviton, Trainium, and Nitro combined — now exceeds a $20 billion annual run rate, growing 100% year-over-year, making AWS the third-largest datacenter chip business behind NVIDIA and AMD.
At an IEEE conference this week, Huawei committed to chip designs with transistor density equivalent to 1.4nm by 2031 — parity with TSMC's projected roadmap — despite ongoing US export controls on advanced semiconductor access. 2

Policy

Google Cloud launched the Gemini Enterprise Agent Platform at Google Cloud Next '26 in Las Vegas — a unified platform for building, orchestrating, and governing AI agents at enterprise scale. 2
The Vatican published a new encyclical on May 25 calling for international AI regulation and warning against uses that undermine human dignity. 6 Separately, US intelligence agencies are reportedly continuing to use Claude in active operations despite the government's recent designation of Anthropic as a national security concern — a designation driven by Claude Mythos's reported cyber capabilities. The Pentagon is running parallel evaluations of OpenAI and Google models as potential replacements. 2
The White House AI executive order remained stalled as of this morning's reporting.

围绕这条内容继续补充观点或上下文。

  • 登录后可发表评论。