The AI map split three ways in a single week: autonomous agents moved inside the everyday apps enterprises already run, a European sovereign-AI bloc consolidated around a $20 billion merger, and cost-efficiency overtook raw capability as the metric that decides who buys what. For technology leaders the question is shifting from “which model is smartest” to “where does it run, what does it cost, and whose law governs the data.”
OpenAI
What happened
OpenAI tuned GPT-5.5 Instant across ChatGPT and the API for more natural, better-paced answers with fewer bloated, bullet-heavy responses, and added in-chat writing and code blocks. Codex gained Computer Use on Windows — it can now see, click, and type inside Windows apps — plus remote continuation from a phone or Mac and new Codex Profiles for token and usage tracking.
What it means for your agentic build
ChatGPT is now a hands-on automation surface for Windows fleets, not just a chat window, with cross-device handoff that suits distributed teams. Pilot Codex on one real Windows workflow, instrument it with the new usage tracking, and validate the remote-continuation model before you widen the rollout.
Anthropic
What happened
Anthropic shipped Claude Opus 4.8, lifting its agentic-coding score from 64.3% to 69.2% and reasoning-with-tools from 54.7% to 57.9%. Claude Code added dynamic workflows that orchestrate tens to hundreds of background agents, alongside Managed Agents with self-hosted sandboxes and MCP tunnels. The company also expanded its PwC alliance and secured a SpaceX compute deal that raised usage limits.
What it means for your agentic build
Claude Code is now a credible engine for large-scale background automation, and the PwC certification push plus higher limits de-risk scaling. Before you fan out hundreds of agents, stand up per-agent cost ceilings and governance so autonomy does not outrun your controls.
Google DeepMind
What happened
Google launched the Gemini 3.5 family, starting with 3.5 Flash — its strongest agentic and coding model yet, beating Gemini 3.1 Pro on hard benchmarks — and cut the Ultra subscription from $250 to $200 a month. Monthly Gemini users reached 900 million. Google also previewed Gemini Spark, a 24/7 personal agent, and Gemini Omni for multimodal video, with 3.5 Pro arriving next month.
What it means for your agentic build
A cheaper, more capable Flash tier resets total-cost-of-ownership math for high-volume chat, summarization, and batch pipelines. Re-run your TCO model against 3.5 Flash pricing, migrate cost-sensitive inference now, and reserve capacity planning for 3.5 Pro when it ships.
Perplexity
What happened
Perplexity’s Computer agent went live inside Microsoft 365 — Word, Excel, PowerPoint, Outlook, and Teams — on May 29, and its API became a full model-agnostic platform spanning agent, search, embeddings, and sandbox endpoints. The company now reports more than 100 million monthly active users. On May 28, CNN sued Perplexity over alleged copying of roughly 17,000 works, adding to suits from the New York Times, Reddit, and Dow Jones.
What it means for your agentic build
Perplexity now competes head-on with Microsoft Copilot inside the productivity tools your teams already live in — but the mounting copyright litigation is real procurement risk. Pilot Computer in one team as an agentic research layer, and require IP-indemnification language before any enterprise-wide standardization.
Meta AI
What happened
Meta launched Muse Spark, the first model from its Superintelligence Labs under chief AI officer Alexandr Wang — and it is proprietary, not open-weight, marking a clear retreat from the open Llama strategy after Llama 4’s weak reception. The pivot coincides with roughly 8,000 layoffs begun May 20 and a planned $115–135 billion in 2026 AI infrastructure spend.
What it means for your agentic build
The open-weight assumption that underpinned many roadmaps is no longer safe to make with Meta. Audit any Llama dependency now and diversify open-weight needs toward Qwen, GLM, DeepSeek, or Cohere, and treat the reorganization as a signal of near-term roadmap volatility.
xAI
What happened
xAI released Grok Build 0.1, billed as its fastest coding model, in public API beta on May 29. It follows Grok 4.3 — with a 1-million-token context window, native video input, and built-in reasoning — plus new Custom Skills and Connectors for SharePoint, Outlook, Google Workspace, Notion, GitHub, and Linear. xAI is also courting Wall Street firms including Apollo and Morgan Stanley to pilot Grok.
What it means for your agentic build
Grok is becoming an integrated coding and automation agent that reaches into the enterprise apps your teams already use. Trial it where you already run SharePoint, Google Workspace, or GitHub, and watch the financial-sector pilots for evidence it can clear regulated-industry bars before you commit.
DeepSeek
What happened
DeepSeek’s V4 line keeps resetting cost curves. At a 1-million-token context, V4-Pro uses only about 27% of V3.2’s compute and 10% of its memory, beating all open models on math and coding and trailing only Gemini 3.1 Pro on world knowledge. Effective June 1, V4-Pro API pricing drops to roughly a quarter of list price as a promotional discount ends.
What it means for your agentic build
DeepSeek is now the cost leader for self-hosted and high-context workloads. Benchmark V4-Pro against your current model for math, coding, and retrieval, but factor China-origin governance and data-residency constraints into any production deployment.
Cohere and Aleph Alpha
What happened
Cohere is acquiring Germany’s Aleph Alpha in a roughly $20 billion all-stock merger endorsed by both the German and Canadian governments, with Schwarz Group committing $600 million to Cohere’s Series E. The deal lands days after Cohere released Command A+, a 218-billion-parameter mixture-of-experts model with native citations, lossless W4A4 quantization, and full Apache 2.0 licensing that runs on two H100s.
What it means for your agentic build
Together these create the clearest on-prem, sovereignty-first option for regulated data that cannot leave your environment. If you run defense, finance, healthcare, or public-sector workloads, pilot Command A+ via Model Vault or North; if you already depend on Aleph Alpha, map the integration path and confirm support continuity early.
This Week’s Structural Trends
Agents go enterprise-native. Perplexity inside Microsoft 365, Codex on Windows, Claude Code’s multi-agent workflows, Mistral’s new Vibe agent, and Grok’s Connectors all push autonomous agents into the apps teams already use. The competition is moving from standalone chatbots to automation embedded in the workflow.
Sovereign AI consolidates into a bloc. The Cohere–Aleph Alpha merger, Mistral’s sovereign-AI partnership with Airbus, and Command A+’s on-prem Apache 2.0 release point to a regulated-data, jurisdiction-controlled alternative forming outside the US hyperscalers — aimed squarely at defense, finance, healthcare, and European public sector.
Cost-efficiency is the new battleground. DeepSeek V4’s compute and memory collapse, Gemini 3.5 Flash’s price cut, and Grok 4.3’s cost-efficient flagship all signal that inference economics, not raw capability, increasingly decide enterprise adoption.
Sources
CNN v. Perplexity: https://www.cnn.com/2026/05/28/media/cnn-sues-perplexity-ai-copyright | OpenAI: https://openai.com/news/ | Anthropic: https://www.anthropic.com/news | Google DeepMind: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/ | Meta: https://venturebeat.com/technology/goodbye-llama-meta-launches-new-proprietary-ai-model-muse-spark | xAI: https://x.ai/news | DeepSeek: https://www.technologyreview.com/2026/04/24/1136422/why-deepseeks-v4-matters/ | Mistral: https://www.cnbc.com/2026/05/28/mistral-arthur-mensch-design-chips-ai-data-centers.html | Cohere: https://venturebeat.com/technology/cohere-cracks-lossless-quantization-and-native-citations-with-first-full-apache-2-0-licensed-open-model-command-a | Aleph Alpha: https://betakit.com/cohere-to-acquire-germanys-aleph-alpha-in-sovereign-ai-play/