Tech • AI • Robotics • Game

ENFR
TodayPlay for freeFury BloxLeaderboardShortsTop StoriesFor youTopicsVideosYT channelsArchivesSearchFavorites

Google Gemini Omni 1.2 Flash leads as Qwen 4 leaks and AI stocks lose $2.5T

AISunday, November 8, 2026· 9 videos

Briefing

Audio player

AI stocks erase $2.5 trillion

A technology-led selloff wiped roughly $2.5 trillion from equity values in a single session before a partial rebound, with the Nasdaq down about 5% at the low. Selling centered on AI infrastructure, semiconductor groups, and other high-beta growth names whose valuations had become tightly linked to continued data-center expansion. Investors were reacting not only to stretched multiples but also to fresh political risk after the U.S. midterms. The result was a sharp split between exposed AI trades and more diversified megacaps such as Apple.

Midterms rattle data-center buildout

The policy shock behind the market reversal was a perception that newly empowered officials may be less willing to grant AI expansion a free pass. In Texas, the incoming governor signaled support for business investment but said future data-center growth would need stronger local protections and clearer rules. That mattered because power, land use, and permitting have become core assumptions in the AI capex story. Investors repriced companies most dependent on aggressive infrastructure rollout.

Google Gemini Omni cuts video costs

Google DeepMind released Gemini Omni 1.2 Flash on November 6, extending clip length from 10 to 15 seconds while cutting price by 30%. At 720p, the model is priced near $0.07 per second, or about $1.05 for a full clip, undercutting several rivals. It also ranked first in human-preference arenas for text-to-video and image-to-video, and third in video editing. The package strengthens Google's commercial case in ads, virtual try-on, and iterative brand content.

GPT Image 2.6 adds GIFs

OpenAI is rolling out GPT Image 2.6 across ChatGPT and the API, with a notable addition: native GIF creation. The update also improves local edit precision, preserving surrounding details when users change only one element of an image. OpenAI paired that with library-style reference handling, letting users save assets and recall them with simple mentions for more consistent characters and branding. The release points to a broader push from one-off image generation toward reusable visual workflows.

Alibaba Qwen 4 leak points higher

Signs are mounting that Alibaba is close to launching the Qwen 4 family, including Qwen 4 Max, Flash, Plus, and a 27B variant. A leaked promotional image indicated 1 million-token context, image support, and tool calling, with reports suggesting an open-weight positioning. Unexplained arena checkpoints under names such as Kira, Enzia, and Meera are being read as preview builds tied to the coming lineup. If confirmed, the release would sharpen competition in coding, web development, and multimodal agents.

Anthropic records 1.9x capability acceleration

An internal-access study by METR inside Anthropic found frontier model progress accelerated after October 2025 to roughly 29 AECI points per year, up from about 15 before that. The reported increase was 1.9x, with a 95% confidence interval of 1.5 to 2.8, reinforcing fears that AI may be starting to speed AI research itself. The strongest unreleased system, Model 4, scored 178.6 on AECI, above Opus 5.5 at 174.5 and Mythos 5.1 at 172.5. Researchers concluded the trend likely represents a bend in the curve rather than a one-time jump.

Hidden model race widens

Fresh identifiers from Anthropic, OpenAI, Alibaba, and Google suggest the unreleased-model race is broadening across coding and multimodal systems. At Anthropic, Claude Donut EAP appears linked to a Silicon High family with five reasoning levels, while Claude Oniiri EAP is associated with Obsidian 3, a model rumored to post strong coding results. OpenAI internal tooling has surfaced codenames including Cactus and Kraken, adding to signs of parallel experimentation. The pattern indicates vendors are testing multiple families and cost-capability tiers before public launches.

Stack Overflow searches for relevance

Stack Overflow has fallen from 207,000 monthly questions at its March 2014 peak to fewer than 1,000 new questions a month, underscoring how conversational AI has reshaped developer behavior. The site remains a major archive, but the social engine that once produced fresh answers at scale has weakened sharply. In response, the company has emphasized data licensing and enterprise offerings over pure community growth. Its trajectory has become one of the clearest measures of how coding assistants are displacing legacy knowledge platforms.

Videos covered

Previous briefings · AI