Claude Mythos Launch Imminent, GPT-5.6 Kindle Alpha in Design Arena, OpenAI Files for IPO, Apple Unveils Siri AI

Multiple major stories converged today across the AI landscape. Anthropic appears to be hours away from releasing Claude Mythos. OpenAI finalized GPT-5.6 under the Kindle Alpha checkpoint and filed for an IPO. Apple introduced Siri AI at WWDC26. Google's Gemini 3.5 Pro leaks surfaced. New open-source models, desktop agents, and research tools also launched.

Anthropic: Claude Mythos Release Could Come Tomorrow

Multiple sources indicate Anthropic is preparing to release the public version of Claude Mythos as early as June 10. New model checkpoints detected over the weekend — Claude Fable 5 and Claude Fruitcake EAP — are believed to be connected to the Mythos family.

Fable 5 is described by leakers as a Mythos-class model with stronger public-facing safety guardrails, potentially serving as the generally available version while raw Mythos remains the underlying frontier model. API references have been appearing across platforms, red-teaming has reportedly completed, and model demos are being finalized.

Polymarket now shows a 92% probability that Anthropic releases Mythos this month, with some predictions specifically pointing to June 10.

Leaked Outputs from Fable 5

Early outputs attributed to Fable 5 show significant capability. A fully functional replica of the game Cut the Rope was generated in a single shot with multiple levels and complete gameplay mechanics. Another output created with low thinking effort still produced strong results, and testers noted a noticeable quality jump between reasoning modes. The model generates between 2,000 and 4,000 lines of code per output without shortcuts or lazy completions.

Tokyo Workshop

Anthropic is hosting a workshop in Tokyo, Japan on June 10. The company has historically used these events to announce new model releases or product updates, adding further weight to the expectation of an imminent launch.

OpenAI: GPT-5.6 Kindle Alpha in Design Arena, IPO Filed

GPT-5.6 Nears Release

The Kindle Alpha checkpoint — previously identified as a release candidate — has now appeared inside Design Arena, suggesting OpenAI is finalizing GPT-5.6 for public deployment. Early comparisons between Kindle Alpha and the earlier Kelpar Alpha checkpoint produced mixed results, with some tests showing Kindle as a slight regression and others showing normal variance. Regardless, Kindle has been selected as the version that will ship as GPT-5.6.

One leaked example demonstrated a zero-shot SVG recreation of an Xbox controller from a single sentence prompt with no tools, indicating strong visual understanding and layout recreation capability.

OpenAI Files for IPO

OpenAI has submitted a confidential S-1 filing with the SEC, initiating the process toward a public offering. The company stated there is no fixed timeline and that it may still operate as a private company for some time, but the filing gives it the option to move forward when conditions are favorable.

This follows Anthropic's IPO filing and SpaceX's upcoming listing. Together, these three companies could produce one of the largest IPO cycles in technology history.

Interactive Charts in ChatGPT

OpenAI also rolled out a small but useful feature: ChatGPT users can now generate interactive charts from data comparisons and analysis directly on web and mobile, turning static tables into visual, interactive graphics.

Google: Gemini 3.5 Pro Leaks Show Partial Progress

Leaked outputs from Google's upcoming Gemini 3.5 Pro have appeared online. The model shows improvement but still exhibits the "laziness" problem that has plagued earlier Pro variants — incomplete generations and shortened responses when asked to build something detailed. Google is actively testing new checkpoints to make the model more competitive against GPT-5.6 and the upcoming Mythos release. Early outputs should be evaluated as unfinished work rather than final quality.

Apple WWDC26: Siri AI and Google Gemini Partnership

Apple introduced Siri AI at its WWDC26 conference, marking the company's most significant AI product announcement. While many of the demonstrated capabilities are already available in ChatGPT and other assistants, Apple's advantage is deep OS-level integration.

Siri AI can dig through messages, emails, photos, and app data to answer contextual questions, plan routes, and take actions across applications without manual navigation. One demo showed Siri finding photos from a trip, filtering by location and specific people, and adding only those photos to a shared family album — all autonomously.

Apple also confirmed a deeper collaboration with Google. Gemini models will be available to Apple developers through the Foundation Model Framework and Xcode, enabling developers to swap between local and cloud inferences and use Gemini-powered coding assistance inside the Xcode IDE.

Nex N2: New Open-Source Agentic Model Family

A new open-source model family called Nex N2 has been released and is not receiving the attention it deserves. These are agentic models — designed not just to think but to act across coding, searching, and tool use within a single reasoning loop.

Key features include:

Adaptive thinking mode: Automatically adjusts reasoning depth based on the task, saving approximately 20% of tokens without performance loss
Coherent thinking: Maintains logical consistency across extended reasoning chains
Strong benchmark performance: Competes with GPT-5.5 and Opus 4.7 on SWE-Bench, Terminal Bench, and GDP Evaluation, beating them in certain categories
Open-weight: Available for local testing and comparison against closed frontier models

Notebook LM: Agentic Upgrade

Google upgraded Notebook LM with a more agentic architecture. The research tool is now powered by Gemini 3.5 Flash and provides better visibility into the AI's reasoning process. Each notebook gains access to a secure cloud computer with over 100 curated software skills for deeper research and complex multi-step workflows.

The most significant addition is the agentic research flow: users can start with loose ideas or questions, and Notebook LM can build out a notebook by finding and adding relevant web sources with user permission. The upgrade is rolling out to Google AI Ultra subscribers.

Kimi for Work and Kimi Code

The team behind Kimi released Kimi for Work, a desktop application for macOS and Windows with several notable agentic features:

Native agent swarm: Users can spawn up to 300 local agents to work on tasks in parallel
Browser use: Agents interact with the web through Kimi's web bridge system for research automation and workflows
Memory system: Retains context across tasks and improves with continued use

Kimi Code, the open-source coding agent, also received a major upgrade with one-line CLI installation, faster startup, support for dragging in videos as coding context, new plugins for stocks and financial reports, MCP protocol support for JetBrains and Zed, and custom workflow hooks.

Summary

Anthropic — Claude Mythos potentially launching June 10; Fable 5 checkpoint shown producing full game clones in single shots; Polymarket at 92% for June release
OpenAI — GPT-5.6 Kindle Alpha confirmed in Design Arena as release candidate; S-1 filed for IPO; interactive charts added to ChatGPT
Google — Gemini 3.5 Pro leaks show improvement but persistent laziness issues
Apple — Siri AI unveiled at WWDC26 with deep OS context; Gemini integration in Xcode
Nex N2 — Open-source agentic model family, adaptive thinking, competitive with GPT-5.5 and Opus 4.7
Notebook LM — Agentic research flow, Gemini 3.5 Flash, cloud computer with 100+ skills
Kimi for Work — Desktop app with 300-agent swarms, browser use, memory system; Kimi Code upgraded with video context and MCP support