Google launched Gemini 3.7 Flash on August 13 1 at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens — half the original Gemini 3.6 Flash rate. AutomationBenchZapier's benchmark measuring whether AI models complete multi-step real-world business workflows from a stated goal. (Zapier’s benchmark for completing multi-step business workflows from a stated goal) jumped from 17.0% on 3.6 Flash to 30.4%, a 79% gain in just three weeks. DeepSWE v1.1An agentic software-engineering benchmark measuring long-horizon autonomous coding task completion on real repositories. (an agentic software-engineering benchmark for long-horizon coding tasks) rose from 49.0% to 65.3%. FrontierCode 1.1 Main improved from 34.4% to 43.6%. Gemini SparkGoogle's 24/7 personal AI agent tier for Pro and Ultra subscribers, running on Gemini models and taking autonomous actions across connected apps. — Google’s 24/7 personal AI agent for Pro and Ultra subscribers in 160 countries — switched to 3.7 Flash on launch day, making the improved model the default runtime for autonomous commerce tasks executed on behalf of users. Google confirmed the same week that the Gemini app surpassed 1 billion monthly active users 2, executing automated tasks across more than 40 connected applications. Google added 14 new third-party app connections to Gemini 3, including Ticketmaster, Zocdoc, Angi, GetYourGuide, and OpenTable for UK users, extending Spark’s autonomous action reach into ticketing, healthcare appointments, and home services.
The three-week gap between 3.6 Flash (July 21, 2026-w30) and 3.7 Flash is the shortest inter-release interval in the Flash series. The Flash family has now shipped three versions in 2026: 3.5 Flash in May (2026-w22), 3.6 Flash in July, and 3.7 Flash this week — each posting a higher AutomationBench score than its predecessor. 3.7 Flash’s result of 30.4% exceeds Claude Opus 5’s published score on the same benchmark at its launch (2026-w30). The 14 new app connections 3 build on Ask Maps food ordering (2026-w32) and AI Mode OTA booking partners (2026-w26), extending agentic reach from restaurants and travel into live events, home repairs, and medical appointments. Spark users can also connect custom Model Context Protocol (MCP, the open standard for agent-to-tool connections) servers, giving operators a programmatic path to make services Spark-accessible without waiting for a first-party integration.
Anthropic published the technical details of its text watermarking system on August 11 4, implementing EU AI Act Article 50The EU AI Act provision requiring generative AI providers to mark AI-generated content in machine-readable formats; enforceable from August 2, 2026. (the provision requiring generative AI providers to mark AI-generated content in machine-readable formats, enforceable from August 2). Anthropic applies the watermark globally across every commercial surface: the Claude Platform API, claude.ai, Claude Code, Claude Cowork, Claude Tag, and models accessed through AWS, Google Cloud, and Microsoft Foundry. The statistical mark travels with text when users copy and paste it and persists through some editing. For files, Anthropic uses C2PACoalition for Content Provenance and Authenticity — an open standard for attaching verifiable provenance metadata to digital files. (the Coalition for Content Provenance and Authenticity open standard for machine-readable provenance metadata). Anthropic joins Google, Meta, OpenAI, and Black Forest Labs in adhering to the EU Transparency Code. Cross-copy persistence distinguishes this approach from metadata-only implementations that strip provenance when text moves to a new document.
Article 50 now applies to every major AI platform serving EU users, covering product copy, recommendation text, and any output that an agent produces on behalf of a merchant or consumer. Google signed the Transparency Code on July 24 (2026-w30) using SynthID across text, images, audio, and video. OpenAI committed to the same code simultaneously. Anthropic’s August 11 disclosure 4 adds Claude to that set. The overlap between the EU transparency requirements and agentic commerce surfaces is direct: agents that write product listings, generate negotiation responses, or draft merchant communications on behalf of operators will embed machine-readable provenance in every output under these obligations. HM Treasury’s July 14 consultation on Modernising Payment Services Regulation — which includes explicit questions on how consent, authentication, and liability rules apply to agentic payments — remains open for responses until October 6, 2026.