The AI Catchup -- June 9, 2026

Welcome back. This week has a genuine headliner: Anthropic put its most capable "Mythos-class" model into general availability for the first time, with a pricing clock you need on your calendar. Below that, OpenAI keeps widening what Codex is for, ChatGPT changes how it remembers you, and Cursor makes canvases a team artifact.

Let us get into it.

This Week in AI

Anthropic launched Claude Fable 5 and Claude Mythos 5 (vendor release June 9), bringing its most capable Mythos-class model to general use for the first time. Fable 5 is the safe-for-general-use version, available today and callable on the Claude API as claude-fable-5. Mythos 5 is the same underlying model with safeguards lifted in some areas, restricted to vetted cyberdefenders and infrastructure providers through Project Glasswing. Both are priced at $10 per million input tokens and $50 per million output tokens -- less than half the price of Claude Mythos Preview, per Anthropic.

Why it matters, and what to act on:

• Mind June 23. From today through June 22, Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost. On June 23, Anthropic says it will remove Fable 5 from those plans and usage will require usage credits, until capacity allows. If your team plans to standardize on Fable 5 inside a subscription, the free window is short -- budget for credits now.
• Some traffic silently falls back to Opus 4.8. New safety classifiers cover cybersecurity, biology/chemistry, and distillation; when one fires, the request is served by Claude Opus 4.8 instead of Fable 5. Anthropic says classifiers trigger in under 5% of sessions on average and that more than 95% of Fable sessions involve no fallback -- but test your sensitive prompt paths for quality shifts on flagged topics.
• 30-day retention is now mandatory for all traffic on Mythos-class models, first- and third-party. Anthropic says it will not use that data to train models or for non-safety purposes, but for regulated or privacy-sensitive workloads this is a material change from standard API handling. Review it before routing production traffic through Fable 5.

The practical read: Fable 5 is pitched at long-horizon, hand-it-off agentic work -- large migrations, multi-step analysis, autonomous runs -- where Anthropic says its lead over Opus 4.8 grows with task difficulty. If your workloads are short and simple, the gap narrows. Mythos 5 is the same model with guardrails removed for a small set of trusted organizations; for everyone else, Fable 5 is the one you can actually deploy today.

Tool Spotlight

OpenAI is reframing Codex from a coding agent into a tool "for every role, tool, and workflow," and it shipped three things at once to back that up (Codex for every role, Sites, and annotations, vendor release June 2). The package: six role-specific plugins bundling popular apps and automated skills, Codex Sites for turning analysis into shareable hosted web apps (in preview for Business and Enterprise), and annotations that now reach documents, spreadsheets, and presentations -- not just code and websites.

The throughline with the rest of Codex's recent cadence is surface expansion. The same week, OpenAI shipped a Build iOS Apps plugin (June 4) that lets Codex mirror the iOS Simulator in the in-app browser and hot-reload SwiftUI previews without leaving Codex, backed by XcodeBuildMCP flows and open-sourced in OpenAI's plugins repo. If you build native iOS, that is the closest Codex has come to an in-loop test surface.

What We Are Watching

OpenAI shipped a new ChatGPT memory architecture built on what it calls "dreaming" (vendor release June 4). Dreaming is a background process that curates memories by referencing chat history without prompting, replacing reliance on a manually curated saved-memories list. OpenAI says it carries context forward better, follows preferences across conversations, and updates memories as time passes.

Why we are watching rather than recommending action: it changes what ChatGPT remembers about you automatically, and it is rolling out to Plus and Pro users in the US first, with Free and international users following. If persistent, self-updating memory is part of how your team uses ChatGPT, this is a behavior change to validate -- not just a feature toggle.

Comparison Reading

If this week's Codex and Cursor moves have you reassessing your stack, the Codex CLI vs Claude Code vs Cursor architecture deep-dive is still the right anchor. The architectural choices each tool made a year ago are exactly what let Codex bolt on Sites, plugins, and annotations as cleanly as it has, and what shapes where Cursor's canvas-sharing fits. Read it alongside the launches above and the picks they affect.

Quick Hits

• Cursor shared canvases (June 4): Cursor now lets you share a canvas with your team via URL -- a link to a live snapshot teammates open read-only in the Cursor Dashboard, so you hand off a working dashboard or report instead of a full chat thread. Available on Pro, Teams, and Enterprise.
• Claude Platform ant CLI (June 2): Anthropic's ant command-line tool exposes every Claude API resource as a subcommand -- Messages requests, model listing, file uploads, version-controlled agents and environments, and a self-hosted Managed Agents worker -- without hand-writing JSON. Claude Code can drive it natively.
• Fable 5 quick facts: addressable as claude-fable-5; $10/M in, $50/M out; free on Pro/Max/Team/seat-based Enterprise through June 22, then usage credits from June 23; Opus 4.8 is the safety fallback; 30-day retention on all Mythos-class traffic.

---

That is it for this issue. The AI Catchup publishes every Tuesday. If you found this useful, subscribe to get it in your inbox.

Until next week -- stay caught up.