Loading...

Skip to main content

AI/ML

Weekly, source-backed updates for AI and machine learning on Google Cloud — anything in scope for the domain — plus notable moves across the wider industry. Newest first.

Week of Oct 12 – Oct 18, 2026​

What changed in Google Cloud​

  • Gemini 4 Argon (gemini-4-argon) is now generally available on Gemini Enterprise Agent Platform, with a unified 1,048,576-token input and output limit, configurable thinking levels and long decodes; long decodes work only on the v1beta1 API, and Gemini 4 Argon Cyber Permissive (gemini-4-argon-cyber-permissive), a version for defensive cybersecurity, is generally available behind the Fairwind Program allowlist. — Gemini Enterprise Agent Platform release notes, Oct 13
  • Gemini 4 Argon is available through the Gemini API at an introductory price of 2permillioninputtokensand2 per million input tokens and 10 per million output tokens; after the introductory period ends on November 30, 2026, the price is 4and4 and 20. — Google DeepMind blog, Oct 13
  • Gemini 4 Argon is generally available in the Gemini Enterprise web app, mobile app and AI developer tools, and AlphaEvolve experiments can use it; in the apps it is off by default until the toggle defaults to on from October 27, 2026, while in AI developer tools it is on by default unless an organization uses a custom model list. — Gemini Enterprise release notes, Oct 13
  • Administrators can now also make Claude Opus 5.5 and Claude Sonnet 5.5 available in the Gemini Enterprise app, following the AI developer tools (October 8); the models are generally available in the global location and the us and eu multi-regions and are charged pay-as-you-go. — Gemini Enterprise release notes, Oct 15
  • DeepSeek-V4.1-Flash is available in preview as a fully managed model in Model Garden, with text and image input and a context window of up to 1M tokens. — Gemini Enterprise Agent Platform release notes, Oct 14
  • Agent Search's stable answer generation model changed from gemini-2.5-flash/answer_gen/v1 to gemini-3.5-flash/answer_gen/v1, so apps pinned to the stable model will see different answers. — Vertex AI Search release notes, Oct 16
  • The Cloud Vision API deprecated its emotion labels: face detection now returns UNKNOWN for all emotion-related labels, and the Video Intelligence API made the same change. — Google Cloud release notes, Oct 15
  • Gemini Enterprise can now show source citations for answers from custom Model Context Protocol data stores, configured with the citation_mappings parameter in the REST API, generally available. — Gemini Enterprise release notes, Oct 16

Week of Oct 4 – Oct 11, 2026​

What changed in Google Cloud​

  • Google announced the Gemini agent at Gemini at Work 2026: a single agent for questions, knowledge work, media and code that plans multi-step work, can spawn sub-agents and coworker agents with their own identities, and picks between Gemini and Anthropic Claude models for each job; the announcement gives no availability stage or date. — Google Cloud blog, Oct 8
  • Industry-specific versions of Gemini for Financial Services and Legal are in preview, with Government, Healthcare and Retail announced as coming soon. — Google Cloud blog, Oct 8
  • New Gemini Code Assist Standard and Enterprise subscriptions can no longer be purchased from October 9, 2026; existing subscriptions auto-renew through the rest of 2026, auto-renewal ends in 2027, and Google points users to Antigravity through Gemini Enterprise. — Gemini Code Assist release notes, Oct 9
  • Gemini Enterprise administrators can now enable Anthropic Claude Opus 5.5 and Claude Sonnet 5.5 in its AI developer tools (Antigravity 2.0, Antigravity CLI and Antigravity for IDEs); third-party models are off by default, run only in the global, us and eu locations, and are billed as consumption that requires overages to be turned on. — Gemini Enterprise release notes, Oct 8
  • The Gemini API deprecated gemini-3.7-flash and gemini-3.5-flash, which are now automatically routed to gemini-3.8-flash and gemini-3.6-flash, and the deep-research-pro-preview-12-2025 agent shuts down on October 23, 2026. — Gemini API changelog, Oct 8
  • Gemini Nano Banana 2.1 (gemini-nano-banana-2.1), an image generation and editing model, is now generally available on Gemini Enterprise Agent Platform. — Gemini Enterprise Agent Platform release notes, Oct 6
  • Anthropic's Claude Haiku 5.5 is available in Model Garden on Gemini Enterprise Agent Platform. — Gemini Enterprise Agent Platform release notes, Oct 7
  • Google DeepMind released EmbeddingGemma 2, an Apache 2.0 open embedding model with 740 million parameters that maps text, code, images, audio and video into one embedding space for on-device search and retrieval. — Google DeepMind blog, Oct 6
  • Model Armor added enhanced prompt injection and jailbreak protection for Workspace content such as emails and documents, on the us and eu multi-region endpoints for templates using filter version v3 or later. — Model Armor release notes, Oct 9
  • Gemini 3.8 Flash is now generally available in Gemini Enterprise in Singapore with in-region data residency and ML processing. — Gemini Enterprise release notes, Oct 8
  • Gemini Enterprise pay-as-you-go usage above quota now also covers the Frontline, EDU and Emerging Market editions. — Gemini Enterprise release notes, Oct 7

Week of Sep 27 – Oct 3, 2026​

What changed in Google Cloud​

  • Gemini Enterprise added federated query mode and Knowledge Catalog integration in preview for AlloyDB for PostgreSQL, BigQuery, Cloud SQL, and Spanner connectors, letting Gemini Enterprise query operational and analytical data in place through the Model Context Protocol using the querying user's own credentials. — Gemini Enterprise release notes, Oct 2
  • Model Armor added template-specific exclusion rules in preview, letting teams configure dictionary and regular-expression rules per template to reduce false positives in its filters. — Model Armor release notes, Sep 28
  • Vertex AI Workbench shipped a bulk security patch remediating critical- and high-severity CVEs in its custom container images. — Vertex AI release notes, Sep 28

Industry​

  • Commentary: Anthropic released Claude Sonnet 5.5, a faster, lower-cost complement to Claude Opus 5.5 that runs roughly 30% faster and typically costs up to 30% less than Sonnet 5 for the same work at unchanged list pricing, and made it available on Google Cloud alongside AWS and Microsoft Azure — worth a look for any team running Claude models on Vertex AI. — Anthropic, Sep 28

Week of Sep 20 – Sep 26, 2026​

What changed in Google Cloud​

Industry​

  • Commentary: Anthropic released Claude Opus 5.5, cutting typical workload cost by about 40% compared to Opus 5 while matching Claude Fable 5.1-level performance on most work, and made it available on Google Cloud alongside AWS and Azure — worth a look for any team on Vertex AI's Claude models weighing a model upgrade. — Anthropic announcement, Sep 22

Week of Sep 14 – Sep 20, 2026​

What changed in Google Cloud​

Week of Sep 6 – Sep 12, 2026​

What changed in Google Cloud​

  • Gemini Enterprise Agent Platform's Computer Use and Shell sandboxes reached general availability, giving agents an isolated browser or Linux shell environment in which to automate tasks. — Gemini Enterprise Agent Platform release notes, Sep 9
  • Gemini Enterprise Agent Platform's Agent Gateway can now enforce VPC Service Controls perimeter rules for agent communications when configured with VPC connectivity in ALL_TRAFFIC egress mode. This support applies only to deployments created after September 8, 2026 that use an agent connectivity template. Agent Gateway can also associate up to two Agent Registry instances, one global and one regional or multi-region. — Gemini Enterprise Agent Platform release notes, Sep 9
  • Gemini Enterprise's pay-as-you-go edition and its AI developer tools are now available to all projects linked to an invoiced Cloud Billing account. — Gemini Enterprise release notes, Sep 10
  • Gemini Enterprise's agent search in Google Cloud Marketplace now supports semantic search, so you can find agents by describing what they do instead of matching exact keywords. — Gemini Enterprise release notes, Sep 8

Industry​

  • Commentary: DeepSeek officially released the DeepSeek-V4.1-Flash model, under the new default model name deepseek-flash, worth a look for teams evaluating cheaper open-weight alternatives alongside Gemini. — DeepSeek API updates, Sep 10