OpenAI GPT-5.5, Distillation Claims, and AI Workflows

Good morning, {{ first_name | AI enthusiasts }}

The AI frontier keeps shifting — and this week the momentum swings back toward OpenAI. The company's GPT-5.5 (codename "Spud") launched with strong benchmark wins and broad availability, arriving as Anthropic faces rate-limit and quality complaints. Below: the key takeaways, context, and practical workflows — plus a special call for reader submissions.

Top stories

OpenAI retakes the frontier with GPT-5.5

OpenAI released GPT-5.5, pitched as a "new class of intelligence." The model posts leading scores across reasoning, agentic tasks, computer use, and coding tests for public models. Several metrics are comparable to Anthropic's Mythos.

• Performance: High marks on reasoning, agents, and coding benchmarks.
• Speed & efficiency: Matches GPT-5.4 speed with improved compute efficiency — reportedly by rewriting GPU code using Codex and 5.5 itself.
• Pricing & availability: API pricing listed at $5/$30 per million input/output tokens. OpenAI positions this as economical for frontier coding workloads. GPT-5.5 is rolling out across ChatGPT plans and Codex Thinking/Pro variants with generous usage allowances.

Why it matters: OpenAI's rapid shipping and strong benchmarks shift sentiment after months of Anthropic momentum.

U.S. flags 'industrial-scale' distillation by Chinese labs

The White House memo accuses Chinese firms of large-scale distillation campaigns targeting U.S. frontier AI outputs. Distillation refers to training smaller models on outputs from frontier models.

• Allegations: Thousands of fake API accounts and jailbreaks to collect outputs.
• Context: Anthropic previously raised private complaints against DeepSeek, Moonshot, and MiniMax. The new memo elevates this to federal concern.
• Response: The Chinese embassy called the claims "pure slander." A related House bill would push distillation offenders toward export restrictions.

Why it matters: The memo reframes how gains may be achieved — via scraping and distillation rather than fresh architecture research.

Claude: build a personalized Morning Brief

Turn daily updates into a single, ranked newspaper-style brief using Claude or Claude Cowork. This workflow pulls from Slack, Notion, Gmail, and Calendar to produce actionable summaries and schedule prep.

1. Connect Slack, Notion, Gmail, and Calendar in Claude Cowork.
2. Prompt example: "Create a static Morning Edition from my Slack, Notion, Gmail, and calendar updates from the last 24 hours. Rank what matters most and format it like a newspaper with top stories, action items, and schedule prep."
3. Review the draft, give feedback, and turn the workflow into a reusable skill.
4. In Claude Cowork, schedule a recurring task to run the skill each morning.

Pro tip: Have other agents gather outside news into a Notion DB, then use the skill as the editor.

Sponsored & featured tools

Agentspan — build durable agents

Agentspan is an open-source framework for building durable agents. An upcoming webinar covers production stacks, making existing frameworks durable (LangGraph, OpenAI Agents SDK, Google ADK), and real-world patterns for resilience.

Optimizely Opal — AI Marketing University

Free live workshop (1 hour/day, Mon–Fri) to build three working AI agents in five days. Target: senior marketers who want to embed AI into workflows and reclaim time.

Research & industry signals

AI's biggest productivity winners are also the most worried

Anthropic's wide survey links higher productivity gains from Claude to greater anxiety about displacement. Key findings:

• Workers whose roles use Claude most report displacement fears at three times the rate of low-use roles.
• Engineers and early-career workers show elevated concern, despite productivity gains like faster tasks and free time.
• The trend suggests anxiety can rise where tools deliver the largest efficiency boosts.

Other notable updates

• Anthropic traced Claude Code quality problems to three bugs and adjusted subscriber usage limits.
• OpenAI launched ChatGPT for Clinicians (free for verified U.S. health workers) with GPT-5.4 achieving top HealthBench Pro scores.
• Meta plans a 10% workforce reduction in May, citing AI efficiency among causes.
• SpaceX is reportedly discussing partnerships with French AI startup Mistral and coding startup Cursor.
• Tencent open-sourced a Hy3 preview — a model from a rebuilt training stack aimed at agentic coding and search agents.

Trending AI tools

• GPT-5.5 — OpenAI's new flagship.
• Ultrareview — Claude Code command for multi-agent code reviews.
• Qwen3.6-27B — Alibaba's 27B model with improved coding performance.

Community workflow spotlight

Each newsletter features a reader workflow. This week Mark M. from Virginia shares a tax-season workflow that saved hours.

Workflow summary:

• Photographed handwritten donation lists and uploaded to Claude.
• Claude transcribed lists, verified charities with the IRS, grouped items, applied fair-market values by condition, and produced professional PDF summaries.
• Outcome: Nearly 400 fully documented deduction items in minutes, defensible for tax purposes.

Reader tip: For anyone donating, use an AI transcription + validation workflow to create defensible records.

Call for submissions

We love hearing how readers implement AI. Share your community workflows here. New: audio submission option and interview flow. Selected workflows may be featured in an upcoming newsletter.

Closing

See you soon,

Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Related topics

This issue also touches adjacent areas like gaming news (for AI-driven NPCs, procedurally generated content, and model-based testing), developer tools, enterprise AI adoption, and geopolitics of model reuse.