Tags
Switch languages mid-conversation and the avatar's lips and expressions adapt with it, across 97 languages in Google's Live Avatar for Gemini 3.8 Live, which is designed to avoid any loss in video fidelity.
Four of the five Anthropic model-and-tier cells ProvenBrief's new retention tracker could verify keep prompts by default, and Claude Fable 5.1 in GitHub Copilot won't run at all unless Anthropic stores them.
OpenAI's Defense Factory blueprint went live September 9, and the July 2026 Hugging Face incident sits inside it as the precedent for putting $1 billion of subsidized Daybreak access behind frontline defenders.
Some teams are pushing 10x their old code volume, a few claim 50x, and Harness field CTO Martin Reynolds says the pull requests just pile up in review. Harness rebuilt its Git repo for agent traffic and added an AI reviewer.
Custom MCP server connections landed in the Business edition of Google's Gemini Enterprise, marked Public Preview. Google says admins can wire in private data, internal tools and MCP-compliant third-party systems.
Microsoft cut its speech-to-text price from $0.36 an hour to $0.10, roughly 72% in five months, and bundled speaker diarization, timestamps and 60 languages into that launch rate.
"1% of my engineers are responsible for 40% of my token spend," Coder CEO Rob Whiteley told The New Stack, and his fix is Agent Relay, which runs Cursor's agents inside customer-controlled workspaces.
Gemini Live added natural voice commands for multi-step tasks across Docs, Sheets, Drive, and the web, so a spoken request can set up work across those tools.
Retailers running shopping agents on Claude see carts up to 35% larger and shoppers 60% more likely to complete a purchase, Anthropic says, and the blueprint behind them is now a repo teams can fork.
Past roughly 30–50 available tools, Claude starts picking the wrong one. Anthropic's tool search, now running on Claude deployments hosted on Azure in Microsoft Foundry, cuts tool definition tokens by over 85%.
Most new open models are drifting toward hybrid Mamba/attention; IBM's Granite 4.2 went the other way with dense, decoder-only reasoning models, and the 8B often lands near the 30B on benchmarks.
Staff open Claude and their MCP connectors are already authorized, because the organization sets access centrally. Anthropic's Enterprise-managed auth reached general availability on August 24 for Team and Enterprise plans.
$35 million in credits goes to patching open source vulnerabilities, Anthropic says, as it puts Mythos 5 into its Claude Security scanner. Users see findings and patches while direct model access stays restricted.
Per Bloomberg, Anthropic plans to give business customers more control over their data on its most capable models, easing a retention policy written against potential cyberattacks. A new safety system is due later this year.
Indirect channels crossed 40% of Anthropic's ARR in 2Q26, per SemiAnalysis, counting Bedrock, Foundry and Gemini Agent Enterprise. That slice doesn't get monetized the same way direct sales do.
OpenAI is inviting ChatGPT Business customers off the waitlist for $100/month Premium seats: 5x the usage of a Standard seat, no five-hour limit, and a weekly quota reset. Early participants can get up to $500 in credits, $100 per Premium seat added.
OpenAI has a new chief revenue officer: Dali Rajic, president and COO of Wiz, replacing Denise Dresser, who lasted nine months. COO Brad Lightcap and Fidji Simo, who ran AGI deployment, have also left in the past month.
Rubrik's CTO told The New Stack that Anthropic's Mythos Preview immediately turned up complex vulnerability chains its usual tools missed. Hiring more reviewers to keep pace was off the table, so Rubrik built a software harness around the model instead.
OpenAI's cyber-specialized models Daybreak Blue and Daybreak Red are live in Amazon Bedrock, so enterprises can run them inside their own AWS environments. Blue covers frontier general models including GPT-5.6 Sol. Access requires an approved Daybreak Access application.
