GenAI Daily - November 13, 2026: Nvidia's Huang Resigns, Google Ships Gemini Decisions API, Baidu Opens ERNIE 5.5 Flash Preview

GenAI Daily - November 13, 2026: Nvidia's Huang Resigns, Google Ships Gemini Decisions API, Baidu Opens ERNIE 5.5 Flash Preview

Top Stories

Nvidia CEO Jensen Huang Resigns, CFO Kress Steps In - UPDATE

Nvidia CEO Jensen Huang Resigns, CFO Kress Steps In - UPDATE

This follows the board probe reported earlier this week.

Nvidia's board announced that Jensen Huang resigned as President, CEO and board member, effective immediately.

CFO Colette Kress is interim CEO while the board searches for a permanent successor, considering internal and external candidates.

Harvey C. Jones was named the board's first-ever chair.

The exit followed a New York Times report that five employees alleged Huang was romantically involved with a subordinate, and an earlier admission of an extramarital affair with a woman U.S. intelligence officials deemed a Chinese spy.

Huang has agreed to testify before Congress ahead of Thanksgiving.

Nvidia says it expects operations, customer commitments and the product roadmap to continue without interruption. The company kept its Nov. 17 fiscal Q3 report, with results around 1:20 p.m. PT and a call at 2 p.m. PT.

Shares fell more than 4% by Thursday's close after plunging earlier in the day.

AMD rose 3%.

CNBC | Nvidia press release | PBS NewsHour

Why it matters: Teams with multi-year GPU commitments should treat Nov. 17 guidance and any permanent-CEO announcement as the signals for roadmap and allocation risk, since leadership is interim and the company has said nothing about changes to plans.

Google Launches Gemini Decisions API, Priced at $0.10 per Million Input Tokens

Google Launches Gemini Decisions API, Priced at $0.10 per Million Input Tokens

Google introduced the Gemini Decisions API on Nov. 11, powered by a new Gemini Decision 1 model that evaluates up to 1,000 questions in a single request.

Developers can check whether a condition is met or pick from a predefined set.

It accepts multimodal inputs with a context window of up to 256,000 tokens.

Google reports 150 to 250 ms end-to-end latency on contexts up to 16,000 tokens, at $0.10 per million input tokens with no output-token charges. Google also says it uses the API in Search to detect adversarial behavior involving MCP tools.

It is a direct answer to OpenAI's Decisions API. OpenAI made that one generally available on 22 October, with up to 400 questions per request. Caveats: each question is evaluated independently, so apps must handle dependencies in code or follow-up requests. The release is in beta on the Gemini Developer API rather than as an established Google Cloud service.

Integrations are rolling out across OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode.

Google Blog | InfoWorld | Channel Insider

Why it matters: Routing, policy checks and guardrail classification no longer need a generative call, and with two major vendors at the same input price, teams can benchmark calibration and latency on their own workloads before committing.

Baidu Opens ERNIE 5.5 Flash Preview API, Apache 2.0 Weights Due Nov. 25

Baidu Opens ERNIE 5.5 Flash Preview API, Apache 2.0 Weights Due Nov. 25

Baidu released ERNIE 5.5 Flash Preview and opened API access through its Qianfan platform. It supports a million-token context window and uses 14 billion of its 280 billion parameters per token.

Qianfan pricing is RMB 1 per million input tokens, RMB 4 per million output tokens, and RMB 0.04 for cached input.

Baidu says it is also free on OpenRouter.

Baidu claims generation speeds above 200 tokens per second, and says weights will be released by November 25.

Treat it as unfinished. Baidu's release page calls the preview an intermediate post-training checkpoint and warns some capabilities may fall short. The licensing matters: the Apache 2.0 promise differs from the Base model, which shipped earlier this month under Baidu's own agreement.

ERNIE release page | TechNode | Baidu press release

Why it matters: If the weights land under Apache 2.0 on schedule, self-hosting teams get a large sparse model with a 1M window and no custom license, but they should verify benchmark and speed claims independently.


Key Developments

Kazakhstan-Linked Group Used Open-Source Hermes Agent to Run Espionage Operation

Kazakhstan-Linked Group Used Open-Source Hermes Agent to Run Espionage Operation

Singapore-based CloudSek reported that a Kazakhstan-tied hacking group used Hermes, an open-source autonomous agent framework, to steal more than 2,000 national security-related documents from Tajikistan. Researchers found it after the hackers left a server exposed to the internet.

CloudSek said the attacker told Hermes what to do, and the agent planned the attacks, wrote the rules, ran them, kept notes and wrote reports.

Hermes was used for reconnaissance from around mid-August, with a break-in to Tajikistan's electronic document system a couple of weeks later.

The agent wrote a custom program to download, organize and chunk the files.

Reuters via U.S. News

Impact: Defenders should expect agent-driven intrusions that run at machine pace and write their own tooling, so detection that depends on human-speed command patterns needs revisiting.

Microsoft and Salesforce Both Put Money and Seats Behind Pilot-to-Production

Microsoft and Salesforce Both Put Money and Seats Behind Pilot-to-Production

On Nov. 9 Microsoft introduced Frontier Accelerate, a Microsoft-wide package covering Azure, Databases, Copilot, Business processes, Security, AI-ready productivity and Marketplace.

It includes FastTrack and Cloud Accelerate Factory deployment help at no additional cost, plus funding for projects from planning to production. Both are subject to eligibility and approval.

Salesforce took a different route. It reports Agentforce Coworker has passed 185,000 users, up 21x in three months, and is adding a $50 per user, per month option.

Coworker for Microsoft Teams, iOS and desktop clients, and a Coworker MCP server are slated for general availability this month.

The announcement doesn't compare the flat fee with Agentforce's credit-based pricing, so buyers will have to run that math. Microsoft is subsidizing deployment while Salesforce is reintroducing a predictable seat price, and both are aimed at stalled rollouts.

Microsoft Cloud Blog | UC Today

Impact: Buyers with stalled pilots should ask for deployment funding in renewals and model flat-seat versus consumption pricing for agent workloads before signing.

Salesforce Details Agents That Retrain Its Embedding Models

Salesforce Details Agents That Retrain Its Embedding Models

Salesforce published SFR-AutoR&D, a self-improving system that trains and improves the embedding models used in enterprise AI. The retrieval layer matters because when an Agentforce agent answers a question grounded in customer data, it reaches through Data Cloud and an embedding model decides what comes back.

In one reported result, the system produced a model with 36% more throughput and 17% lower latency, with almost no loss in quality.

Salesforce

Impact: Retrieval cost and latency are now being tuned by automated pipelines, so Agentforce customers should watch for embedding-model updates and re-test retrieval quality after each one.


Product Launches

Meta Muse Ember

Meta Muse Ember

Meta's smallest Muse model is available in Muse Code and the Meta Model API as muse-ember-1.0, at $0.10 per million input tokens, $0.004 cached input and $0.50 output, with a 1M-token context and up to 512K output tokens.

List prices match Haiku 5.5 and GPT-6 Luna, so the case rests on cost per task: Ember scores 36 on the Artificial Analysis Intelligence Index, level with Haiku 5.5 at its highest setting, for roughly $0.14 per task against $0.41. Watch the long-prompt tier: above 200,000 input tokens it costs $0.30 in and $1.50 out.

Meta | Martin Cid Magazine

Why it matters: With list prices matching rival small models, Ember's appeal depends on its lower cost per task, so teams should test it on their own workloads, especially if prompts will exceed the 200,000-token tier where pricing rises.

Salesforce Agentforce Coworker Expansion

Salesforce Agentforce Coworker Expansion

Coworker is moving into Microsoft Teams and native apps. General availability this month covers Teams, the iOS app, Mac and Windows desktop clients, the MCP server for external connectivity, and a free trial for Flex Credits customers.

It adds a seat-priced option at $50 per user per month for teams that want predictable costs.

Wazzup

Why it matters: Availability inside Teams and on mobile and desktop puts Coworker where employees already work, and the seat-priced option gives budget owners a predictable alternative to credit-based billing.

Funding & Deals

Daylight Raises $80M Series B

Daylight Raises $80M Series B

New York and Tel Aviv-based Daylight sells managed detection and response run largely by AI agents, with human security engineers customizing the agents to each client's systems. Its Aura platform also covers AI platforms organizations use, offering protection against prompt injection, risky MCPs and rogue AI-driven activity.

The company says revenue grew more than 40-fold over the past year, driven by large enterprises and AI-native companies, and Globes put the valuation at about $400 million; the company did not disclose one. Total funding is now $120 million. Led by Growth Equity at Goldman Sachs Alternatives.

Calcalist | SecurityWeek

Why it matters: Security buyers are increasingly paying for AI-run detection and for protection against threats aimed at AI systems themselves, and Daylight's reported growth suggests demand for both is rising.

Tomorrow's Watch List

Tomorrow's Watch List

  • Nov. 17: Nvidia fiscal Q3 results and call, the first under interim CEO Colette Kress.
  • Before Thanksgiving: Huang's agreed congressional testimony.
  • Nov. 25: Baidu's deadline for ERNIE 5.5 Flash weights under Apache 2.0.
  • November: Agentforce Coworker general availability on Teams, mobile and desktop.

Related reading: Check out this week's [Deep Insights analysis] for strategic context on these developments.


GenAI Daily is curated by rAInvent. For in-depth analysis, see the rAInvent articles.

Read more