<p><strong>TLDR:</strong> Microsoft and OpenAI have restructured their partnership, ending Microsoft’s exclusive license and letting OpenAI serve customers on any cloud through 2032. Anthropic is now charging Claude Code Pro users extra to access Opus, following a rough stretch that included a regression in Claude Code quality and a Mythos preview breach. GitHub is moving Copilot to token-based billing on June 1, driven by the compute costs of agentic workflows. China blocked Meta’s $2B acquisition of Manus, setting a precedent that could affect any Chinese-founded AI startup considering a US deal.</p>
<hr>
<h2>Microsoft and OpenAI end exclusivity</h2>
<p>The renegotiated partnership replaces Microsoft’s indefinite exclusive license with a non-exclusive arrangement running through 2032. OpenAI can now serve customers on AWS, Google Cloud, or anywhere else — Azure gets first-ship rights, but that’s it. Microsoft stops paying revenue share to OpenAI; OpenAI continues paying Microsoft through 2030 with a cap. The AGI clause, which gave Microsoft interpretive power over when AGI had been reached, is gone entirely — replaced with language stating payments continue “independent of OpenAI’s technology progress.” Microsoft shares fell roughly 3% on the news. The immediate trigger was OpenAI’s February deal with Amazon, which conflicted with Microsoft’s prior exclusive rights and needed to be unblocked.</p>
<h2>Anthropic paywalls Opus access in Claude Code</h2>
<p>Claude Code Pro subscribers now need to purchase extra credits to use Opus, the most capable model in Anthropic’s lineup. Previously it was included in the plan. This follows another recent restriction blocking new Pro subscribers from accessing Claude Code at all. Community speculation points to Opus consuming Pro quotas too fast under agentic workloads. The timing is rough: Anthropic has also been dealing with a Claude Code quality regression it acknowledged and fixed, a Mythos preview breach, and rate limits pushing some users toward Codex. Whether the extra credit purchase is a one-time unlock or recurring cost remains unclear.</p>
<h2>GitHub moves Copilot to usage-based billing</h2>
<p>Starting June 1, Copilot switches from premium request quotas to a token-based AI credits model. Base subscription prices stay the same across Pro, Business, and Enterprise tiers, but usage beyond the monthly credit allotment costs extra. GitHub also removed the fallback to cheaper models when limits are hit. The driver is straightforward: flat-fee pricing stopped making sense once Copilot became an agentic platform running multi-step coding sessions. Enterprise customers get pooled credits and spending controls. Copilot code review will also start consuming GitHub Actions minutes on the same date, which is worth flagging for teams running private repos.</p>
<h2>China blocks Meta’s $2B Manus acquisition</h2>
<p>China’s NDRC formally ordered Meta to unwind its acquisition of Manus, the agentic AI startup co-founded by Chinese nationals and incorporated in Singapore. The ruling came after a four-month review that included exit bans on Manus co-founders. Around 100 Manus employees had already moved into Meta’s Singapore offices. The case forced a novel regulatory question — whether an AI team constitutes a technology export — and China has since formalized a policy requiring government approval before Chinese tech companies accept US capital. The precedent affects any Chinese-founded AI startup regardless of where it’s incorporated.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Kimi K2.6 released:</strong> Moonshot’s 1T-parameter MoE open-weight model with 32B active parameters claims open-source SOTA on SWE-Bench Pro (58.6) and supports 4,000+ tool calls and 300 parallel sub-agents via “Claw Groups,” with day-0 support on vLLM and OpenRouter.</li>
<li><strong>OpenAI models coming to AWS Bedrock:</strong> OpenAI models are arriving on Bedrock in the coming weeks, with Amazon and OpenAI leaders hosting a joint livestream today on agentic AI.</li>
<li><strong>Ineffable Intelligence raises $1.1B:</strong> David Silver’s post-DeepMind lab, focused on RL-based learning without human data, closed a round led by Sequoia and Lightspeed at a $5.1B valuation, with Nvidia contributing at least $250M.</li>
<li><strong>73 malicious extensions on Open VSX:</strong> The Socket Research Team identified 73 extensions tied to the GlassWorm supply chain campaign, targeting VS Code, Cursor, Windsurf, and VSCodium — at least six are already active. Rotate secrets if you installed any.</li>
<li><strong>GitHub Copilot cloud agent 20% faster:</strong> Optimized runner environments via Actions custom images cut startup time by over 20%, following a 50% improvement shipped in March.</li>
<li><strong>Gemma 4 technical details:</strong> Google DeepMind’s Cassidy Hardin detailed the architecture — four models including a 26B MoE (first in the Gemma family, 3.8B active parameters) and a 31B dense model ranked third on the global AI arena leaderboard, all Apache 2.0.</li>
<li><strong>AI hiring tools pick AI resumes 95-97% of the time:</strong> A study found AI-powered screening tools heavily favor AI-written resumes, raising obvious questions about what these tools are actually selecting for.</li>
<li><strong>Grok Build expected next week:</strong> xAI is reportedly launching a direct competitor to Claude Code and Codex, with no technical details yet.</li>
<li><strong>Devin for Terminal launched:</strong> Cognition’s local shell agent brings Devin’s capabilities directly into your terminal, with cloud handoff for workloads that outgrow local capacity.</li>
<li><strong>Google A2Family announced:</strong> Google open-sourced a suite of agent interoperability protocols — A2A (donated to Linux Foundation), A2UI, AP2, and UCP — with the ADK available in Python, TypeScript, Go, and Java.</li>
<li><strong>DORA data on AI velocity:</strong> The 2025 DORA report found teams with heavy AI tool adoption are shipping 19% slower despite individuals creating 98% more PRs, with review queue overload as the primary bottleneck.</li>
<li><strong>OpenAI smartphone in development:</strong> Analyst Ming-Chi Kuo reports OpenAI is co-designing an agent-first phone with Qualcomm and MediaTek, targeting 300-400M annual shipments by 2028 — a number that would exceed iPhone volumes, from a company that has never shipped hardware.</li>
<li><strong>580+ Google employees urge Pichai to reject Pentagon classified AI deal:</strong> Signatories include 20+ directors and senior DeepMind researchers, citing inability to monitor or limit AI use on air-gapped classified networks.</li>
</ul>


AI coding assistants, code generation tools, and vibe coding workflows.

Agentic Digest

A packed week in AI developer tooling: Microsoft and OpenAI restructured their partnership, ending exclusivity and letting OpenAI serve customers on any cloud through 2032. Anthropic began charging Claude Code Pro users extra to access Opus, citing heavy compute consumption under agentic workloads. GitHub is switching Copilot to token-based billing on June 1, removing flat-fee premium requests. China's NDRC blocked Meta's $2B acquisition of Manus, setting a regulatory precedent for Chinese-founded AI startups. Notable items include Kimi K2.6's 1T-parameter open-weight model, OpenAI models coming to AWS Bedrock, a 73-extension supply chain attack on VS Code and forks, DORA data showing AI-heavy teams shipping 19% slower, and xAI's Grok Build expected next week as a Claude Code competitor.

Microsoft and OpenAI rewrite their deal, Anthropic quietly paywalls Opus in Claude Code

<p>they are using other ai model to give or write proper answer in same mdel where there own model is wrost</p>
