Curated AI news from the last 24 hours — models, moves, and industry shifts.
Anthropic is launching a new enterprise AI services company backed by Blackstone, Goldman Sachs, Hellman & Friedman, and others including General Atlantic, Apollo, and Sequoia. Each anchor investor commits roughly $300M. Rather than acting as a traditional consultancy, the firm will embed Anthropic-trained engineers directly inside mid-sized companies to redesign workflows around Claude — putting it in direct competition with McKinsey and Accenture in the fast-growing corporate AI transformation market.
→ CNBC coverageAnthropic's most capable GA model ships with substantially improved vision (higher-resolution image understanding), stronger performance on the hardest software engineering tasks, and better long-running agentic consistency. Available on the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry at unchanged pricing: $5 / $25 per million input/output tokens. A new Claude Security public beta — scanning code for vulnerabilities and generating fixes — is also live for Enterprise customers.
→ anthropic.comGPT-5.5 Instant replaces GPT-5.3 Instant as ChatGPT's default model. The standout new capability: the model can reference past conversations, uploaded files, and connected Gmail to deliver more personalised answers — available to Plus and Pro subscribers on the web. The full GPT-5.5 (released April 23) remains available via API for developers, priced at current GPT-5 tier rates.
→ openai.comxAI introduced SuperGrok Heavy, a new subscription tier giving access to the full Grok Heavy model alongside significantly higher rate limits. The launch follows Elon Musk's congressional testimony (April 30) that xAI trained Grok partly on OpenAI models — a disclosure that has added legal complexity to the ongoing OpenAI vs. Musk litigation. X also reportedly rebuilt its advertising platform with AI baked into the stack, tightening the xAI-to-monetisation loop.
→ TechCrunch on Grok/OpenAI testimonyNIST's Center for AI Standards and Innovation (CAISI) announced new agreements with Google DeepMind, Microsoft, and xAI allowing the U.S. government to evaluate frontier models — including classified environments — before public launch and after deployment. CAISI has already completed 40+ evaluations. The expansion follows Anthropic's powerful Mythos model raising White House concern about AI cybersecurity risks, though Anthropic itself was excluded from the new Pentagon AI deals over disagreements about military-use guardrails.
→ Nextgov / FCWThe Information reports Anthropic has signed a landmark agreement committing to $200B in Google Cloud spending over the next five years — one of the largest cloud-commitment deals ever. Separately, Anthropic and Google/Broadcom struck a deal in April for multiple gigawatts of TPU capacity starting in 2027, signalling that Anthropic is aggressively locking in compute well ahead of its next generation of models.
→ U.S. News (The Information)Sam Altman has discussed spinning out OpenAI's robotics and consumer hardware units into independently funded subsidiaries operating under an Alphabet-style holding structure. The move would give those divisions faster capital access and operational independence while keeping model development at the core. OpenAI is simultaneously launching its own enterprise AI services joint venture, racing Anthropic's newly announced Blackstone/Goldman venture for the same mid-market territory.
→ TechCrunchA notable May 2026 benchmark finding: Claude Opus 4.7 running inside Cursor's harness outperforms the same model in Claude Code's native harness by nearly four percentage points on functionality benchmarks. Claude Code remains agent-first (you describe, AI drives), while Cursor stays IDE-first (you drive, AI assists). With Opus 4.7, multi-model routing in Cursor — also covering GPT-5 series and Gemini — is pushing developers toward a "use both" workflow depending on the task.
→ SitePoint comparisonThe week's dominant theme: enterprise AI deployment moving from pilot to production at scale. Anthropic's $1.5B services JV and OpenAI's parallel move mark the shift from "which model?" to "who installs it in your company?" Meanwhile, NIST's expanded pre-release testing agreements signal that governments are no longer just watching — they're building infrastructure to evaluate frontier models before they ship.