AI Product Updates Daily — May 30, 2026

A lighter day on new model launches, but busy underneath: OpenAI retired two model families and updated a third, Gemini Spark went live for the first time, Anthropic quietly shipped two new products for different ends of the market, and Cognition's Devin crossed $492M in annualized revenue on the way to a $1B raise. The day's activity also included the first Nvidia-powered Windows PCs announced ahead of Computex, a YouTube feed feature that accepts plain-text prompts, and Google's ISO audit of Copilot expanding to cover Copilot Studio.

GPT-5.5 Instant gets a quality refresh. OpenAI shipped a new version of GPT-5.5 Instant on May 30, fixing what its research lead Michelle Pokrass described as the model being "too bullet-pilled" — meaning it was over-optimized toward lists and defensive responses. The update improves response style, readability, sycophancy scores, factuality, and multilingual performance. 1

One change attached to this update has immediate UX impact: Canvas has been removed from GPT-5.5 Instant and GPT-5.5 Thinking. Writing and coding now happen in-chat through writing blocks and code blocks. Paid users retain access to Canvas through legacy model paths until those are sunset.

Two model retirement dates confirmed. Tied to the GPT-5.5 Instant update, OpenAI published sunset dates for two older models: 2

API access timelines are separate. For teams running production workloads on either model, the June date gives roughly four weeks.

ChatGPT adds personal finance tools via Plaid. A preview for ChatGPT Pro users in the US lets people connect bank accounts, credit cards, savings accounts, and loans. ChatGPT reads balances and transaction data to surface spending trends and cash flow patterns. The connection is read-only — no transfers or payments. The feature runs on a partnership with Plaid. 3

ChatGPT table of contents for long conversations. A smaller UX update: ChatGPT now auto-generates a navigation table of contents once a thread reaches a certain length, letting users jump directly to earlier sections without scrolling. 4

OpenAI publishes Frontier Governance Framework. On May 29, OpenAI released a public document mapping its internal safety and security practices to California's Transparency in Frontier AI Act (TFAIA) and the EU AI Act's Code of Practice for General Purpose AI. Coverage spans cyber offense, CBRN risks, harmful manipulation, and loss-of-control scenarios. This is the first time OpenAI has formalized compliance mapping in a publicly accessible governance document. 5

Gemini Spark goes live in the US. One week after its I/O 2026 debut, Gemini Spark launched in beta on May 29 for Google AI Ultra subscribers (minimum $100/month). 6

Spark runs on Gemini Flash 3.5, built on Google's Antigravity platform, and operates as a background agent — not a chatbot you open and close. It can book flights, build CRM lead lists from Gmail data, track pricing differences across vendors, and run tasks even when your device is off. It appears as a new tab in the Gemini web sidebar and as a distinct layer on Android and iOS between search chats and Daily Brief.

At launch, Spark connects to Canva, OpenTable, and Instacart outside the Google ecosystem. Adobe, Uber, Spotify, and Booking.com are listed as upcoming partners. Google plans to add custom sub-agents and local browser control in future updates.

Google Search adds Preferred Sources to AI Overviews and AI Mode. Google expanded its Preferred Sources feature — previously limited to standard search — into AI Overviews and AI Mode. Users can add websites by name or URL; those sources appear with a visible badge in AI-generated answers rather than being buried below general links. Google says chosen sources are clicked twice as often. More than 345,000 sources have been added by users so far. 7

YouTube rolls out AI prompt-designed feeds. YouTube is releasing a feature to US users (logged in, with watch and search history enabled) that lets them describe a preferred feed in a text prompt. The custom feed pins to the top of the homepage for 30 days, can be updated at any time, and coexists with the regular feed. Sample prompts Google offers: "give me something different beyond my usual feed" and "help me unwind after work with guided meditations under 10 minutes." One limit: only one custom feed active at a time. 8

Claude for Small Business launches. Anthropic shipped a new product tier targeting SMBs, connecting Claude to six platforms: Intuit QuickBooks (payroll, monthly close, cash flow, tax prep), PayPal (invoicing, disputes, refunds), HubSpot (lead triage, campaign attribution), Canva (content creation and publishing), DocuSign (contract workflows), and both Google Workspace and Microsoft 365. 9

The design is closer to a workflow layer than a chatbot: users connect existing software stacks, select tasks, and review Claude's outputs before they execute. A free AI Fluency for Small Business course is launching with PayPal as part of the rollout.

Claude Opus 4.8 pricing note. For teams reviewing this week's Claude Opus 4.8 launch: the model maintains previous pricing tiers but adds a Fast Mode at $10/$50 per million tokens (input/output), down from $30/$150 in the previous premium tier — a 3x reduction for high-speed access. 10

Microsoft 365 Copilot clears ISO 42001 audit; Copilot Studio added to scope. Microsoft published results of a March 2026 ISO/IEC 42001 surveillance audit for Microsoft 365 Copilot: zero non-conformities, zero improvement observations. The scope now includes Copilot Studio, meaning the governance standard covers custom agents and connected workflows — not just the base chat interface. This is the second consecutive year of recertification. 11

ISO/IEC 42001 certifies that an organization's AI management processes are documented and governed — not that individual model outputs are safe in every deployment. Enterprise teams still need to test tenant-specific permissions and agent behavior before broad rollout.

Cognition raises $1B+ at $26B valuation. The AI coding agent startup behind Devin closed a round led by Lux Capital, General Catalyst, and 8VC. Annualized revenue is reported at $492M, growing at roughly 50% per month on the enterprise side. 12

Clients include Mercedes-Benz, NASA, Goldman Sachs, Santander, and Nubank. Enterprise usage is up 10x since early 2026. Cognition plans to use the funding to expand paid deployment in large engineering teams, with Claude Code, Codex Symphony, and Cursor as direct competitors.

The valuation has moved from $10.2B in September 2025 to $26B in under a year.

Nvidia and Microsoft tease Nvidia-powered Windows PCs ahead of Computex. On May 30, Reuters and Axios reported that Microsoft, Dell, and Nvidia will unveil the first Windows computers running on Nvidia chips next week at Computex (Taiwan) and Microsoft Build (San Francisco). 1314

Microsoft's first AI PC push built around Qualcomm and Intel ran into adoption headwinds. Nvidia's arrival changes the framing — the company's brand recognition in performance computing gives the category a clearer pitch for developers and power users.

Groq pivots fully to inference cloud. After selling its chip technology to Nvidia in a $20B licensing deal in December 2025, the remaining Groq team is raising $650M to operate as a pure AI inference cloud service with no hardware business. Existing investors Disruptive and Infinitum have backstopped the full amount. New leadership is Adam Winter (CEO) and Matt Eng (CFO). 15

ByteDance weighs $70B AI infrastructure spend. Bloomberg reported ByteDance is considering capital expenditures of up to $70B in 2026 for AI data centers, up from $25B in 2025. The company earned $50B in profit in 2025 and could largely self-fund the buildout. Its Doubao chatbot has 300M+ monthly users in China. 16

Azure AI Metrics Advisor retired May 18. Microsoft retired Azure AI Metrics Advisor as of May 18, 2026. Recommended alternatives are Azure Monitor (for time-series anomaly detection) and similar native Azure 3P tools. 17

DALL-E API sunset (reminder). DALL-E 2 and DALL-E 3 were removed from the OpenAI API on May 12. Teams still calling those endpoints should migrate to GPT-Image models. 18