It’s Topical
← The AI Builder Roundup archive
The AI Builder Roundup

Claude gets a browser, Gemini gets ears

Your AI stack just picked up a browser, better ears, longer context, persistent memory, and cheaper images. Pick one workflow that has been annoyingly manual and give it a fresh test this week.

In today’s edition

  • Claude’s built-in Cowork browser can navigate sites and fill forms from a separate desktop side panel.
  • Gemini 3.5 Transcribe adds smart dictation features and is available through the Gemini API.
  • GLM-5.3-Flash brings multimodal, long-context agent work to OpenRouter with discounted launch pricing.
  • Perplexity’s Brain gives Computer a structured, continually updated memory system for files, sessions, and sources.
  • Muse Image is available through OpenRouter at $0.01 per image for production-scale creation and editing.

Claude puts a browser inside Cowork

Claude now opens a browser in Cowork’s side panel when a task needs a website, then navigates, fills forms, and completes the job.

  • The browser is built into the Cowork desktop app, so there is nothing to install.
  • It stays separate from your own browser and logins.
  • The built-in browser is rolling out over the next week to paid desktop plans.
  • Claude in Chrome is also generally available for paid plans if you prefer working in your signed-in browser.

What it means Test browser-based workflows such as form entry and site research in Cowork, while choosing the separate browser or Chrome option based on login and access needs.

Source: claudeai

Gemini 3.5 Transcribe targets real voice work

Google introduced Gemini 3.5 Transcribe, a speech-to-text model built for precise, smart dictation across apps and devices.

  • It is designed to handle self-corrections and remove filler words for cleaner formatted text.
  • The model understands natural intent and supports custom vocabulary, multi-speaker identification, and more than 85 languages.
  • It includes function calling and real-time streaming support.
  • Builders can use it through the Gemini API via Google AI Studio, Antigravity, and the Gemini Enterprise Agent Platform public preview.

What it means Use it to prototype voice agents, meeting capture, and dictation flows where clean transcripts, custom terms, and streaming responses matter more than raw audio-to-text conversion.

Source: GoogleAI

GLM-5.3-Flash lands with a 1M-token window

OpenRouter revealed that Ox Alpha is Z.ai’s GLM-5.3-Flash, the first native multimodal model in the GLM-5 series.

  • OpenRouter says Ox Alpha processed more than 20 trillion tokens in six days.
  • GLM-5.3-Flash is positioned for efficient coding and long-horizon agent tasks.
  • It supports a 1M-token context window and up to 131K output tokens.
  • Launch pricing is 50% off through September 9 at 16:00 UTC: $0.075/M input, $0.25/M output, and $0.015/M cached input.

What it means Benchmark GLM-5.3-Flash on document-heavy coding and agent workflows now, especially if its launch pricing can lower the cost of long-context experiments.

Source: OpenRouter

Perplexity gives Computer a working memory

Perplexity’s Brain is a self-improving memory system for Computer that turns sessions, files, and sources into a structured knowledge wiki.

  • Brain uses durable memory storage, foreground agents that answer queries, and background agents that update memory.
  • It makes a user’s history available without loading all of it into every prompt.
  • The memory is stored as linked Markdown files, with claims connected to the supporting sessions and files.
  • Perplexity reports new evals improved correctness by 9.3 points, currentness by 8.0, and recall by 8.9 while using 15% fewer tokens.

What it means Treat durable, evidence-linked memory as a core agent pattern: retrieve context on demand, preserve the source trail, and let scoped background jobs keep it current.

Source: perplexity_ai

Muse Image hits OpenRouter at one cent

Muse Image from Meta is now available on OpenRouter, with search-grounded image generation and editing priced for production volume.

  • The listed price is $0.01 per image.
  • It can create and edit complex images.
  • Edits preserve the parts of an image you do not change.
  • OpenRouter positions the price for high-volume creative work.

What it means Run the numbers on product catalogs, ad variants, and other image-heavy workflows where one-cent generation can make automated creative production viable.

Source: OpenRouter

More from The AI Builder Roundup →