The AI Catchup Newsletter
A weekly roundup of AI tools, practices, and what matters, curated so you can stay current without the noise.
Past Editions
The AI Catchup - October 13, 2026
Google released Gemini 4 Argon to paid API developers, Antigravity on Ultra, and the Gemini app on Pro and Ultra. It leads the Vals Index at 68.9%, writes up to 1M output tokens, and costs $2 and $10 per million tokens through November 30. Also: ChatGPT Intelligent UI, Claude Dashboards and Motion, and OpenAI's Decisions API.
The AI Catchup - October 7, 2026
Anthropic released Claude Haiku 5.5, its fastest model and the first Haiku with effort levels. It scores 72.4% on OSWorld against Haiku 4.5's 15.7% and costs $0.10 and $0.50 per million tokens under 100K tokens, 90% less. Sonnet 5.5 cache reads drop to $0.10, Max and Team plans get monthly API credits, Claude Code gets TypeScript mods, and Cursor documented Remote Control for local agents.
The AI Catchup - September 29, 2026
OpenAI's DevDay brought GPT-6.1 Sol, which edges past GPT-6 Astra's best DeepSWE score at a fifth of Astra's token price, dots, always-on ChatGPT agents with their own cloud computer, and Pro 500, a $500 plan with 25x Plus usage and Astra Ultrafast, up to 8x faster in Codex. Pro 200 reportedly drops to 10x.
The AI Catchup - September 22, 2026
Claude Opus 5.5 and OpenAI's GPT-6 Sol and Luna all shipped September 22. Opus 5.5 leads Anthropic's benchmark table at $4 and $20 per million tokens, and at medium effort beats GPT-6 Astra's best knowledge-work score for about a fifth of the cost. Sol halves to $2 and $10 and wins high-volume automation. Luna costs $0.10 and $0.50.
The AI Catchup - September 1, 2026
Claude Fable 5.1 is out today as claude-fable-5-1. Same $10 in and $50 out per million tokens as Fable 5, but cache reads drop 75% to $0.25, which Anthropic says makes typical workloads about 25% cheaper and agentic ones up to 45% cheaper. Cyber safeguards fire about 60% less in Claude Code, zero data retention is back for eligible enterprises, and Mythos 5.1 stays behind two vetting programs.
The AI Catchup - August 28, 2026
Cursor Cloud Agents now wake on a PR, a Slack thread, or a schedule and hold a /goal instead of needing a prompt per request. Claude Code sessions can be started from a phone, not just continued there. ChatGPT Voice can launch and steer Codex. Claude Code resumes itself after a limit reset. Ox Alpha has a name.
The AI Catchup - August 12, 2026
SpaceXAI dropped Grok 4.6 into Cursor and Grok Build with 2x included usage for the first week. Claude Code sessions can now message each other on Mac and Linux. OpenAI shipped Agent Plugins, expanded Luna to Free and Go, and opened Daybreak Red with GPT-5.6-Cyber.
The AI Catchup - July 22, 2026
Cursor doubled usage limits on every individual and teams plan, three days after Anthropic extended its Claude Code promo to August 19. Claude Code Desktop now runs your iOS app in a live simulator pane and taps through it to verify its own work. Google shipped Gemini 3.6 Flash and 3.5 Flash-Lite with aggressive agent pricing. Plus: Codex 0.145.0 and GPT-Red.
The AI Catchup -- July 15, 2026
OpenAI split ChatGPT into three agents -- Chat, Work, and Codex -- with Work as a new long-running agent for research and finished deliverables, on every desktop plan including Free. Plus: Cursor put GPT-5.6 Sol, Terra, and Luna in the model picker, shipped durable Side chats, and Anthropic launched Reflect.
The AI Catchup -- July 10, 2026
OpenAI made GPT-5.6 generally available -- Sol, Terra, and Luna are live across ChatGPT, Codex, and the API with published pricing, new benchmark numbers, and a predictable prompt-caching model. Plus: Claude Cowork is coming to mobile and web, Anthropic raised API rate limits, and Claude Code Artifacts reached Pro and Max.
The AI Catchup -- June 30, 2026
Anthropic made Claude Sonnet 5 the new default tier -- its most agentic Sonnet yet, with introductory pricing through August 31 and the `claude-sonnet-5` API ID. Then forward: Cursor put cloud agents on your iPhone, the Claude desktop app reached Linux, Codex shipped least-privilege permission profiles, and Anthropic opened Claude Science in public beta.
The AI Catchup -- June 17, 2026
One week after Anthropic put Mythos-class capability into general use, a US government export-control directive forced it to suspend Claude Fable 5 and Mythos 5 for every customer. If you built on `claude-fable-5`, fail over now. Then forward: Cursor wants to host your code with Origin, Bugbot got ~3x faster with a new `/review`, and Anthropic paused the Agent SDK monthly credit.
The AI Catchup -- June 9, 2026
Anthropic just made a Mythos-class model generally available as Claude Fable 5 -- callable today as `claude-fable-5` at $10/M in and $50/M out -- while keeping the same model with safeguards lifted, Claude Mythos 5, locked to vetted partners. Mind June 23: free subscription access ends and Fable 5 moves to usage credits. Plus Codex goes past coding, ChatGPT's 'dreaming' memory gets a new architecture, and Cursor canvases you can share by URL.
The AI Catchup -- May 19, 2026
Claude Code 2.1.142 is a deceptively impactful release: the new `claude agents` flag set finally makes background session dispatch configurable per run, fast mode now defaults to Opus 4.7 (with a documented 4.6 override), and a long-standing bug that silently capped remote MCP tool calls at 60 seconds is fixed. If you are running agent dashboards or remote MCP servers, upgrade week.
The AI Catchup -- May 12, 2026
Two back-to-back Codex CLI drops (0.129 and 0.130) turn the app-server into something you can program remotely, while Claude Code 2.1.133 puts worktree base defaults back where they belong and routes the entire MCP OAuth flow through proxy/mTLS. Plus: Google's Gemini Interactions API has a hard May 20 / June 8 sunset that breaks older SDK clients.
The AI Catchup -- May 5, 2026
Two big agent-runtime drops this fortnight: Cursor put its agent into an `npm install`-able SDK, and Codex CLI 0.128.0 added persisted `/goal` workflows for Ralph-style loops that don't stop until done. Plus: Warp open-sourced its client under AGPL with OpenAI as founding sponsor, and Cursor 3.2 made multitasked, multi-root agent work the default flow.
The AI Catchup -- April 15, 2026
This week was Anthropic's coming-out party for orchestrated coding work -- Routines, a rebuilt desktop app, and Thariq's deep dive on managing the 1M context window. Plus: Warp goes universal, Cursor draws you a UI, and Ramp shows what 99 percent AI adoption actually looks like.
The AI Catchup -- March 4, 2026
This week: MCP servers are everywhere, the Cursor vs Claude Code debate heats up, and we share our guide to building your own MCP server from scratch.