AI Daily

Google opens Gemini 4 Carbon to paying developers and Ultra users

2026-11-10 · New

Google expanded Carbon beyond its earlier specialist program to paid APIs, enterprises and Google AI Ultra users. Public-model API input/output pricing is temporarily $2/$10 per million tokens, rising to $4/$20 after the introductory period ends November 30. Developers can use AI Studio and Android Studio access paths; the Fairwind cyber-defense program still requires an application. Antigravity’s accompanying /swarm multi-agent feature remains a preview, distinct from model availability.

Google: Gemini 4 Carbon availability

Google: Carbon developer access

Meituan releases LongCat-2.5 weights under MIT

2026-11-10 · New

Following the November 2 model announcement, Meituan released LongCat-2.5 weights on November 10 under MIT. The MoE model has 1.6 trillion total and 48 billion active parameters, with a one-million-token context window. The new development is downloadable weights for deployment. Meituan describes W4A4 four-bit inference and domestic-accelerator optimization; memory, throughput and quality still need configuration-specific validation rather than assuming company benchmarks are independent tests.

Meituan: LongCat-2.5 open weights

LongCat-2.5 model card and weights

Meta launches Muse Ember with input pricing from $0.10 per million tokens

2026-11-09 · Recent

Meta made Muse Ember available in Muse Code and Meta Model API as muse-ember-1.0. It supports a one-million-token context window. Input/output pricing is $0.10/$0.50 per million tokens for prompts up to 200,000 tokens, and $0.30/$1.50 for longer prompts. The small model targets agent subtasks such as tool use and extraction, adding a lower-priced option for multi-model workflows. This is a hosted release; a lower token price does not guarantee a lower total cost for every task.

Meta: Muse Ember

LG opens free API testing for K-EXAONE 2.5

2026-11-09 · Recent

LG AI Research introduced the K-EXAONE 2.5 technical validation model, offering free API testing for developers, startups and enterprise research, with further work on coding and agent tasks. Teams can evaluate their own tasks and integration workflows. Reported evaluations are company-provided; technical-validation access does not establish a production service guarantee or unrestricted commercial-use permission.

LG: K-EXAONE 2.5 technical validation model

Google extends Gemini Intelligence to Pixel 10

2026-11-10 · New

Google’s November Pixel Drop starts a phased Gemini Intelligence rollout to the Pixel 10 family, including voice-input refinement and multi-step tasks. Gemini multi-step tasks also expand to Japan, the UK and India. Existing phones gain agent capabilities through software updates. Delivery continues over several weeks and depends on device and region; features described as forthcoming are not treated as already available.

Google: November Pixel Drop

Google and Artificial Analysis introduce a 200-task multimodal work benchmark

2026-11-10 · New

Google and Artificial Analysis introduced MKA, using 200 tasks spanning ten industries and 83 occupations to evaluate knowledge work involving documents, charts and other multimodal materials. A shared toolset and execution harness assess both factual accuracy and presentation, adding a deliverable-oriented comparison for model selection. Results remain specific to these tasks and conditions; they do not establish replacement of the occupations represented.

Google and Artificial Analysis: MKA benchmark

Gemini Carbon · Google

4 · 2026-11-10 · stable

This issue: broader paid API and Ultra access; /swarm is a separate preview.

Google: Gemini 4 Carbon availability

Google: Carbon developer access

Checked 2026-11-11T08:25:02.431691+00:00

LongCat · Meituan

2.5 · 2026-11-10 · stable

This issue: weights released under MIT; distinct from the November 2 model announcement.

Meituan: LongCat-2.5 open weights

LongCat-2.5 model card and weights

Checked 2026-11-11T08:25:02.431691+00:00

Muse Ember · Meta

1.0 · 2026-11-09 · stable

Recent verified baseline: hosted through Muse Code and the API as muse-ember-1.0.

Meta: Muse Ember

Checked 2026-11-11T08:25:02.431691+00:00

TSMC board approves $40.9267 billion in capital appropriations

2026-11-10 · New

TSMC’s board approved approximately $40.9267 billion in capital appropriations for advanced nodes, packaging, mature and specialty capacity, fab facilities and 2027 R&D-related spending. It separately approved investments of up to $500 million each in Anthropic and OpenAI. The resolutions authorize capacity and R&D investment; they are neither annual capex guidance nor money already spent. The equity investments are approved ceilings, not completed transactions.

TSMC November 10 board resolutions

NVIDIA presents GR00T 2 for joint action and future-video prediction

2026-11-09 · Recent

At CoRL, NVIDIA presented the GR00T 2 world-action model, combining future-video prediction with robot control and demonstrating research on transferring tasks from human-action videos to physical robots. The approach aims to reduce task-by-task teleoperation data collection. Current evidence consists of research experiments and demonstrations, not general-purpose robot deliveries. The research page schedules code availability for December, so reproducibility materials are still pending.

NVIDIA: CoRL robotics and simulation developments

GR00T 2 research and code plans

EIA forecasts US generation growth of 2.7% in 2026 and 1.2% in 2027

2026-11-10 · New

In its November outlook, EIA forecasts US electric-power-sector generation of almost 4.39 trillion kWh in 2026, up 2.7%, followed by 1.2% growth in 2027. It identifies commercial demand and summer cooling as 2026 drivers. This updates the power-supply context for data-center expansion, but the figures cover the whole sector and cannot all be attributed to AI. Model inputs were finalized November 5; the forecast does not specifically incorporate later market events and is not installed-capacity data.

EIA: November 2026 Short-Term Energy Outlook

US Senate advances energy-permitting reform by 73–25

2026-11-09 · Recent

The US Senate advanced the BAAJA permitting package by 73 votes to 25. The committee says it addresses environmental-review and energy laws and includes provisions intended to make prospective data centers bear their own power-related costs. Data-center developers need to track the final permitting and cost-allocation provisions. This is a procedural advance; amendments, final votes and subsequent legislative steps remain pending, so it does not establish new rules already in force.

US Senate committee: permitting bill advances

CoreWeave reports $3.661 billion Q3 revenue and a $709 million net loss

2026-11-10 · New

CoreWeave reported Q3 revenue of $3.661 billion versus $1.365 billion a year earlier, operating income of $207 million and a net loss of $709 million. Its quarter-end revenue-backlog metric reached $132.5 billion. The results show both cloud-compute demand growth and financing pressure. Backlog includes remaining performance obligations and other committed amounts; revenue recognition still depends on delivery and service availability, so it cannot be added to realized quarterly revenue.

CoreWeave third-quarter results

Oura agrees to buy Counsel Health for physician-supervised AI care

2026-11-10 · New

Oura signed a definitive agreement to acquire Counsel Health, building on their existing partnership to integrate physician-supervised AI care. The company expects closing near year-end. The service connects continuous wearable context with clinical care. Physicians need user permission to access Oura data, and licensed doctors retain responsibility for diagnosis and prescriptions. The deal is signed, while fuller app integration remains forthcoming.

Oura agrees to acquire Counsel Health