TLDR: Anthropic quietly moved Claude Code to the $100/month Max plan, triggering immediate backlash and a wave of users hunting for alternatives. Meanwhile, Kimi K2.6 landed at the top of multiple open-weight benchmarks, putting real pressure on frontier pricing. Google Cloud Next dominated the enterprise news cycle with a full rebrand of Vertex AI and a $750M partner fund. The Amazon-Anthropic deal context keeps getting bigger — $30B annualized revenue, $800B valuation floated by late-stage VCs.
<hr>
<h2>Anthropic moves Claude Code to $100/month</h2>
The $20 Pro plan no longer includes Claude Code access. Users now need the $100/month Max plan, which Anthropic initially described as a 2% pricing test before the full pricing page changed. The backlash was immediate — searches for “how to cancel Claude” spiked overnight, and workarounds using NVIDIA’s free API tier started circulating within hours. This lands at an awkward moment: Anthropic is reportedly in early IPO discussions and sitting on $30B in annualized revenue, which makes the optics of squeezing the $20 tier harder to defend.
<h2>Kimi K2.6 takes the top open-weight spot</h2>
Moonshot’s Kimi K2.6 hit #2 overall in Code Arena, #1 in Vision Arena, and #1 in Document Arena — all as an open-weight model. Design Arena put it in the same performance band as Claude Opus 4.7 while establishing a new price-to-preference frontier. The community reaction is pointed: Kimi K2.6 is being credited as one of the few things keeping frontier labs from raising prices further. At roughly 1/10th the cost of Claude Opus 4.6 with comparable performance on many tasks, it’s a real alternative for teams that got burned by today’s pricing news.
<h2>Google Cloud Next: Vertex AI becomes Gemini Enterprise Agent Platform</h2>
Google rebranded Vertex AI and consolidated Agentspace into Gemini Enterprise, announcing ADK v1.0, managed MCP servers across Google Cloud services, and the A2A protocol now at v1.2 in production at 150 organizations. The $750M partner fund targets Accenture, Deloitte, KPMG, and PwC to push Gemini as the default enterprise AI platform. Ironwood TPU is now generally available at 4.6 petaFLOPS per chip, and Google previewed eighth-gen TPUs split into separate training (TPU 8t) and inference (TPU 8i) chips — the first time Google has purpose-built separate silicon for each workload. Anthropic is the anchor customer for both generations.
<h2>The Anthropic-Amazon deal keeps getting larger</h2>
Amazon’s total potential investment in Anthropic is now around $33B, with Anthropic committing to spend over $100B on AWS over the next decade. Annualized revenue reportedly hit $30B in early April 2026 — up from $9B at end of 2025 — with Claude Code alone accounting for $2.5B of that. Late-stage VCs are reportedly offering capital at an $800B valuation, more than double the $380B from two months ago. The circular structure is worth noting: Amazon invests, Anthropic spends it back on Amazon infrastructure. Whether that’s a good trade depends entirely on whether the growth trajectory holds.
<hr>
<h2>Also notable</h2>
<ul>
<li>Roo Code pivots away from IDEs: The VS Code extension shuts down May 15, 2026, replaced by a cloud-based agent called Roomote that runs tasks across GitHub, Linear, and Slack without IDE interaction.</li>
<li>Claude Code /ultrareview: A research preview that runs a fleet of bug-hunting agents in the cloud, with findings landing in the terminal — no pricing details yet.</li>
<li>Cursor integrates with Slack: Mention @Cursor in a thread and it uses channel context to open a PR, with real-time streaming updates visible in Slack.</li>
<li>Zed ships parallel agents: Multiple AI agents can now run simultaneously in a single window, with a new Thread sidebar for managing and isolating each agent’s file access.</li>
<li>OpenAI workspace agents: Shared agents in ChatGPT that handle long-running workflows across tools — the same underlying implementation as their open-sourced Codex workspace agent.</li>
<li>Anthropic Mythos unauthorized access: A private Discord group guessed the restricted model’s URL on launch day using familiarity with Anthropic’s URL conventions, accessing a model withheld from public release due to its ability to autonomously discover zero-day vulnerabilities.</li>
<li>Firefox 150 patches 271 Mythos-discovered bugs: Mozilla shipped fixes for vulnerabilities found by Claude Mythos Preview, framing AI-assisted discovery as ultimately good for defenders.</li>
<li>Shopify token usage inflection: A chart of token usage across all Shopify technical staff shows a sharp change in slope starting December 2025, with percentile deltas widening despite an unlimited token budget being in place the entire time.</li>
<li>OpenAI ChatGPT ad model shifts to CPC: The initial $60 CPM collapsed to $25 within ten weeks; OpenAI pivoted to $3-5 per click bids and cut minimum advertiser spend from $250K to $50K.</li>
<li>Gemini on air-gapped servers: Cirrascale partnered with Google Cloud to offer Gemini as a fully private on-premises appliance where model weights live only in volatile memory — power off, model gone.</li>
<li>OpenAI Privacy Filter on HuggingFace: A lightweight open-source bidirectional token-classification model for PII detection with a 128k context window.</li>
<li>SpaceX S-1 contradicts Musk: The pre-IPO filing warns orbital AI data centers involve unproven technologies and may never achieve commercial viability, three months after Musk called space-based AI a no-brainer.</li>
<li>Florida opens criminal probe into OpenAI: The first criminal investigation of an AI company over an alleged role in a mass shooting, based on chat logs from the April 2025 FSU attack.</li>
<li>Meta tracking employee keystrokes: The Model Capability Initiative captures mouse movements, clicks, keystrokes, and screenshots from US employees to train AI agents on how humans navigate software interfaces.</li>
<li>MCP SDK hits 300M downloads/month: The protocol is moving faster than the governance tooling around it — gRPC transport from Google, enterprise auth, and task primitives are all in progress but the security control plane gap remains unfilled.</li>
</ul>

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

Agentic Digest

Anthropic quietly removed Claude Code from its $20/month Pro plan, requiring the $100/month Max plan instead, triggering immediate user backlash and searches for alternatives. Moonshot's Kimi K2.6 topped multiple open-weight benchmarks at roughly 1/10th the cost of Claude Opus 4.6, putting pressure on frontier pricing. Google Cloud Next saw Vertex AI rebranded as Gemini Enterprise Agent Platform with ADK v1.0, managed MCP servers, and a $750M partner fund. The Amazon-Anthropic deal has grown to ~$33B total investment with Anthropic committing $100B in AWS spend over a decade, and annualized revenue reportedly hitting $30B. Additional notable items include Roo Code shutting down its VS Code extension, Cursor integrating with Slack, Zed shipping parallel agents, Claude Mythos discovering 271 Firefox bugs, and MCP SDK reaching 300M downloads/month.

Anthropic pulls Claude Code from $20 plan, Kimi K2.6 tops open-weight leaderboards

<blockquote>
Anthropic is reportedly in early IPO discussions and sitting on $30B in annualized revenue, which makes the optics of squeezing the $20 tier harder to defend.
</blockquote>
It’s as if they’re leaking money and $30B in revenue doesn’t come close to covering costs.

I still see Claude Code in the $17/$20 Pro Plan. Pricing page just not changed/refreshed yet, or Anthropic change of mind?

Tried Kimi K2.6 the other day. Its really really good!
With the claude news i guess the new meta is opencode with kimi.
Or what do you think?