AI Daily

Google launches Gemini 4 Hydrogen at $1.50 per million input tokens

2026-11-12 · New

Gemini 4 Hydrogen reaches AI Studio, the Gemini API, app and Search. Across its one-million-token context, pricing is $1.50 per million input tokens and $7.50 per million output tokens. Antigravity adds Orchestrator to switch between Carbon and Hydrogen within a task. Hydrogen access there requires Pro or Ultra; free app users receive automatic model selection.

Google: Gemini 4 Hydrogen

MiniMax releases H3 Image weights for generation and multi-image editing

2026-11-10 · Recent

MiniMax H3 Image unifies text-to-image and editing with up to 16 reference images in one checkpoint. Its model card describes native outputs of about 16 megapixels, with official weight files available. These are weights under a custom license: it excludes the US, UK, EU and South Korea. Hosted services or annualized revenue above $1 million require separate written authorization. Availability is not unrestricted commercial permission.

MiniMax H3 Image official model card

MiniMax H3 Image community license

ElevenLabs releases Scribe v3 and Live with realtime speaker labeling

2026-11-09 · 7-day watch

November 9 release notes confirm Scribe v3 and Scribe v3 Live in the speech-to-text API for recorded and realtime audio respectively. Live adds speaker identification and end-of-turn prediction. Live accepts up to 10,000 keyterms for specialist vocabulary. The v2 interfaces remain available, allowing teams to test their own recordings and multi-speaker conversations before migrating.

ElevenLabs: Scribe v3

ElevenLabs November 9 release notes

ElevenLabs SDK 2.77.0 adds agent workflows as code

2026-11-09 · 7-day watch

ElevenLabs Python and JavaScript SDKs 2.77.0 add a Workflow builder for conversation, tool, transfer and end steps with conditional routing. Workflows can be versioned as JSON while remaining editable visually. Teams can review configuration changes before deploying through create or update interfaces; authoring alone does not make a workflow live.

ElevenLabs November 9 release notes

ChatGPT Enterprise joins AWS Marketplace with eligible commitment spending

2026-11-09 · 7-day watch

OpenAI listed ChatGPT Enterprise on AWS Marketplace, enabling purchasing through an existing AWS billing relationship and application of eligible spending toward AWS commitments. This adds an enterprise procurement route with the same features and updates as direct purchasing. Customer eligibility and commitment treatment still need confirmation; it is distinct from buying model inference through Bedrock.

OpenAI: ChatGPT Enterprise on AWS Marketplace

Gemini · Google

4 Hydrogen · 2026-11-12 · stable

This issue: flat pricing across 1M context; free app access uses automatic selection.

Google: Gemini 4 Hydrogen

Checked 2026-11-13T05:23:00.116010+00:00

Agent Sandbox · Google Cloud

GA · 2026-11-12 · stable

This issue: GKE isolation component GA; Agent Gateway separately remains in preview.

GKE November 12 release notes

Checked 2026-11-13T05:23:00.116010+00:00

H3 Image · MiniMax

Open weights · 2026-11-10 · stable

Recent verified baseline: weights available under territorial and commercial restrictions.

MiniMax H3 Image official model card

MiniMax H3 Image community license

Checked 2026-11-13T05:23:00.116010+00:00

Scribe · ElevenLabs

v3 / v3 Live · 2026-11-09 · stable

Recent verified baseline: recorded/realtime transcription APIs; v2 remains available.

ElevenLabs November 9 release notes

Checked 2026-11-13T05:23:00.116010+00:00

Sandisk introduces iNAND CL EM181 with an established e.MMC interface

2026-11-10 · Recent

Sandisk introduced the iNAND CL EM181 embedded flash drive at electronica 2026, with 64–256GB capacities for physical-AI and IoT devices such as cameras and gateways, plus data-center logging. Retaining e.MMC 5.1A helps long-lifecycle devices move to newer NAND with fewer host-design changes. The announcement establishes a product introduction and exhibition, without a realized volume-shipment figure.

Sandisk: iNAND CL EM181

MBZUAI releases PAN 2 to simulate scenes from textual actions

2026-11-11 · Recent

MBZUAI’s Institute of Foundation Models released PAN 2: a language model plans timed actions, then a video diffusion model simulates subsequent scenes. The institute announced video weights and training/inference code under Apache 2.0. It offers pre-action simulation for robotics research. The institute still identifies long-term world consistency as a challenge. This is physical-reasoning and simulation research, rather than evidence of deployed robots.

MBZUAI: PAN 2 world model

GKE Agent Sandbox becomes generally available; Gateway remains a preview

2026-11-12 · New

Google made Agent Sandbox generally available across supported GKE versions for isolated, stateful, single-replica agent workloads, with no additional charge for the component. The same-day Agent Gateway release is a preview limited to the Rapid channel, specified versions and two US/European regions. Isolation and traffic management have different release stages; compute charges remain separate.

GKE November 12 release notes

Google Cloud launches Organization Policy MCP and natural-language constraints

2026-11-12 · New

Google Cloud announced general availability of Organization Policy MCP and Gemini-powered natural-language constraint authoring. Agents can help inspect policies, configuration drift and exemptions; administrators can draft custom rules conversationally. Gemini can also generate compliant and non-compliant test resources for dry runs before enforcement. The tools reduce authoring work; administrators still need to check constraint scope and exemptions.

Google Cloud: Organization Policy MCP and natural-language authoring

Australia proposes mandatory frontier-AI evaluations and a dedicated regulator

2026-11-10 · Recent

Australia plans AI Standards legislation requiring frontier-AI companies operating locally to undergo independent evaluation, report incidents rapidly and publish safety frameworks, alongside a proposed dedicated regulator. Developers can prepare evaluation and reporting processes, but details and consultation remain pending. Proposed penalties and obligations are not yet enacted requirements.

Australia: proposed AI safety legislation

Jensen Huang resigns as NVIDIA CEO and director; Colette Kress takes interim role

2026-11-12 · New

NVIDIA’s board announced Jensen Huang’s immediate resignation as president, CEO and director. Finance chief Colette Kress becomes interim CEO while retaining her CFO and executive vice-president roles. The board is searching for a permanent successor and appointed Harvey C. Jones as chair. Customers can track delivery and roadmap disclosures through the transition; operational continuity remains the board’s stated expectation.

NVIDIA: CEO transition

Vercel reports decision models at 26% of October requests and 0.53% of estimated spend

2026-11-09 · 7-day watch

Vercel’s November report says decision models accounted for 26% of October AI Gateway requests but only 0.53% of spending estimated at public list prices. Nearly a third of paying teams used them. Classification, scoring and choice tasks now show measurable usage relevant to procurement. The sample covers Vercel’s gateway only: request share is not industry market share, and estimated spend is not suppliers’ realized revenue.

Vercel: November AI Gateway usage report