The AI Brief

OpenAI added WebMCP support to ChatGPT's desktop browser app on August 25. OpenAI added WebMCP support to ChatGPT's desktop browser; AI agents can now call structured tools on compatible sites instead of scraping interfaces. The feature works with GPT-5.6 Sol and Terra, while Luna currently has WebMCP disabled, and is not available in Enterprise or Edu workspaces.

Millions of Shopify storefronts are already WebMCP-enabled, with Expedia, Instacart, and Target among the companies experimenting with the standard. Shopify confirmed WebMCP tools were live on every Liquid storefront; the very next day, Cloudflare introduced a developer preview that adds a WebMCP bridge at the network edge. WebMCP lets a site define named functions with descriptions and typed inputs; an agent can call those functions inside an existing browser session, where they run in the page's JavaScript and share the user's session.

OpenAI launched a 10-day WebMCP Challenge with $3,000 per winner and prizes from Shopify, Google Chrome, Netlify, Cloudflare, Vercel, and Render. The challenge accepts submissions through September 3. The standard is experimental and adoption outside commerce is nascent; the contest is an early test of whether sites will publish agent-facing actions instead of relying only on interfaces designed for human clicks. Separately, ChatGPT Work also gained webhook-triggered scheduled tasks this week: ChatGPT added webhook-triggered scheduled tasks in Work, plus shared tasks and flexible limits for Free users, and expanded Work's browser to complete tasks on signed-in websites, with secure logins, password manager support, and confirmation before consequential actions. Supported webhooks include new Gmail messages, Slack channel messages, and GitHub pull request activity.

Gemini 3.5 Pro is still unreleased as of August 23, 2026, more than three months after Google announced it at I/O on May 19, having now missed three targets, June, mid-July, and early August, with no model ID, no pricing, and no launch date. As of today, August 30, the count reaches 103 days. Google DeepMind still lists it as "coming soon."

Bloomberg reported in July that Pro is months behind schedule, tied to coding performance that fell short of Google's internal bar, with no new date given; earlier reporting also said Google rebuilt Pro from a new base model rather than the earlier 2.5 Pro one. The best Gemini model callable today is not a Pro model at all, it is Gemini 3.7 Flash, which went GA on August 13.

The gap is now structurally visible in Google's own API changelog: Gemini 3.6 Flash, released to stable GA, features improved token efficiency and code and agentic planning capabilities at a lower price point than 3.5 Flash. Google shipped a downstream Flash variant before its flagship Pro tier shipped at all, an inversion with no clear precedent at the lab. The prior coverage high-water mark was day 97 (Vol. I, No. 96, August 26).

Anthropic launched Claude Sonnet 5 in June with introductory pricing of $2 per million input tokens and $10 per million output tokens, explicitly framed as temporary through August 31. Anthropic announced on August 10 that the introductory pricing for Claude Sonnet 5 is now permanent. When Sonnet 5 launched in June, Anthropic set the rate at $2 per million input tokens and $10 per million output tokens, framed explicitly as introductory pricing through August 31. Standard pricing of $3/$15 per million tokens was scheduled to kick in from September 1. That increase will not happen.

The freeze matters most because of the compounding tokenizer dynamic. Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text. Claude Sonnet 5 runs on this new tokenizer; a team migrating a workload from an older Sonnet model will not just see a higher sticker price; the token count for identical inputs and outputs will likely climb too, compounding the rate increase in practice rather than sitting beside it.

The operative picture today: the September 1 rate-card event has been neutralized by Anthropic's August 10 announcement. Teams that budgeted for the increase and reallocated to alternative models or reduced call volume may now find themselves over-hedged. The previously scheduled increase to $3/$15 per million tokens on September 1 will not occur; Claude Sonnet 5 has become the default workhorse model for AI agent deployments, sitting at the sweet spot of capability and cost for multi-step tasks. First covered as a pending risk in Vol. I, No. 87 (Aug. 10).

Disclosure: Claude, which generates this brief, is built by Anthropic.

On August 30, 2026, OpenAI retired the official DALL·E GPT in ChatGPT, encouraging users to download any images they want to keep before then and directing image creation to ChatGPT Images going forward. User-created GPTs with image generation enabled are not affected.

The DALL·E GPT was the dedicated image-generation wrapper inside ChatGPT's GPT Store, distinct from the native ChatGPT Images capability that OpenAI has been expanding. Its removal follows the retirement of o3 from ChatGPT on August 26, GPT-4.5 from ChatGPT in June, the ChatGPT Atlas browser shutdown on August 9, and the Assistants API hard shutdown on August 26. The pattern reflects OpenAI's deliberate consolidation away from the plugin-and-GPT-Store architecture toward a unified agentic surface built around the GPT-5.x model family and ChatGPT Work.

OpenAI also announced ChatGPT for Academic Researchers, offering faculty and postdoctoral researchers twelve months of complimentary access to a dedicated ChatGPT workspace for teams of up to five members, with business data protections and ChatGPT Pro-level usage limits. That program, launched alongside the DALL·E GPT retirement, signals a simultaneous push into institutional adoption as consumer-facing legacy products wind down.

Google released Gemini 3.6 Flash to stable general availability on August 13, noted in the Gemini API changelog without a dedicated launch post. Gemini 3.6 Flash features improved token efficiency and code and agentic planning capabilities at a lower price point than 3.5 Flash, resolving developer feedback around output verbosity. Google also moved Gemini 3.5 Flash-Lite to GA simultaneously. Gemini 3.6 Flash and Gemini 3.5 Flash-Lite reached stable, production-ready GA; the existing gemini-omni-flash-preview endpoint will be deprecated on September 30, 2026.

The release creates an unusual lineup: Google's most recently shipped production model is Gemini 3.6 Flash, a second-generation Flash release, while the Pro tier that nominally sits above it, Gemini 3.5 Pro, has no model ID, no pricing, and no API entry. The official Google DeepMind models page lists Gemini 3.1 Pro as the current Pro model and carries a "3.5 Pro coming soon" marker; the model is delayed, not cancelled.

Also in the Google API changelog: the gemini-robotics-er-1.6-preview model will be shut down on August 31, 2026 , a low-profile deprecation that removes a robotics evaluation model one day after today's edition. Google separately shipped Gemini 3.5 Transcribe and Gemini 3.5 Transcribe Live to GA, covering high-accuracy non-streaming and low-latency bidirectional streaming speech-to-text across 85-plus languages. The speech models represent Google's most consistent GA cadence of August: three models shipped quietly while its flagship reasoning model stayed in limbo.