Tech • AI • Robotics • Game

VIDEO
ENFR
TodayPlay for freeFury BloxLeaderboardShortsTop StoriesFor youTopicsVideosYT channelsArchivesSearchFavorites

Anthropic misses lofty bar as Meta drops 816B Muse Spark Open

AISaturday, October 17, 2026· 17 videos

Briefing

Audio player

Anthropic growth still jolts markets

Anthropic reported Q3 revenue of $16.8 billion, framed as an annualized run rate of about $67.2 billion, confirming extraordinary scale but not surpassing the market's most aggressive expectations. The result landed in an already fragile tape for AI equities, where investors have been questioning whether revenue growth can keep pace with valuations built on repeated hypergrowth. CoreWeave was cited as down roughly 50% over the past month, illustrating how broad the derating has become. A key disclosure that model distillation accounted for about 5% to 7% of revenue suggested the issue is material, but smaller than some had feared.

Meta opens Muse Spark Open

Meta released Muse Spark Open, an open-weight multimodal model with 816 billion total parameters, 42 billion active, and a 1 million-token context window. The model is positioned for coding, software engineering, AI agents, and long-horizon reasoning across text, images, video, and audio. Reports said it can run locally on very high-end systems, including large-memory Apple Ultra configurations, narrowing the gap between cloud frontier models and local control. Early coding tests portrayed it as competitive in practical generation tasks, including building a playable browser game.

Google broadens Gemini 4 Argon

Google expanded access to Gemini 4 Argon for paying Google AI Pro and Ultra users in the Gemini app and Search AI mode. The company is pitching Argon as more direct in its answers, stronger at coding, and better at handling longer jobs with progress updates. Alongside the model, Tasks lets Gemini act across Google properties and more than 40 external apps including Instacart and OpenTable. Availability appears uneven, however, with plan details suggesting the feature may remain limited to Ultra and selected countries for now.

Claude Fable 5.5 nears release

Developers spotted an Azure entry labeled cloudfable 555, fueling expectations that Claude Fable 5.5 is close to launch. Leaked specifications point to a 1 million-token context window, 128,000-token maximum output, and pricing of $10 per million input tokens and $50 per million output tokens. That profile would target enterprise use cases such as massive document analysis, persistent coding agents, and extended multi-step workflows. Separate reports also suggested Anthropic is quietly expanding faster and cheaper access to Opus 5.5, intensifying pressure across the frontier model market.

Claude records workflows into skills

Claude's desktop app now includes a screen-recorded skill creation feature inside Customization. Users can demonstrate a recurring task by sharing the screen, narrating steps, and letting the app observe interface changes, cursor movement, and screenshots. The system then turns that demonstration into a reusable command, lowering the barrier for workflow automation without requiring carefully structured written prompts. The feature is desktop-only and signals a broader shift from chat interfaces toward teachable agent behavior.

OpenAI breach raises espionage fears

A reported FBI case involving an alleged listening device in a high-level meeting room at OpenAI has sharpened concerns about AI labs as strategic intelligence targets. The suspect was described as a 35-year-old Saudi engineer, and the matter was treated as more than a routine internal security lapse because the room hosted senior executives, government officials, and major partners. Reports said internal monitoring and model-related traces helped identify the suspect after the device was discovered. The episode underlined a widening belief that frontier labs now sit closer to critical national infrastructure than to ordinary software startups.

Starshield disruption hits Ukraine drones

Ukrainian forces reported degraded drone links on the night of October 13-14 once aircraft exceeded roughly 110 to 120 km/h, affecting interceptor operations against Shahed attacks. Ordinary Starlink service reportedly continued, directing attention instead to Starshield, the military-grade network. The slowdown matters because slower defensive drones struggle against Shaheds flying around 150 km/h, while newer jet-powered variants can reach about 400 km/h. The dispute also highlighted the unresolved power of private infrastructure over battlefield capability, with responsibility for restrictions publicly contested between Elon Musk, SpaceX, and the U.S. government.

Tesla Roadster show overshadows product

Tesla's Roadster reveal generated attention with claims of sub-one-second 0 to 60 mph acceleration and SpaceX-inspired cold-gas thrusters. But the most consequential detail was timing: production is not expected until 2028, years after early reservations opened. Reservation terms remain steep, including $5,000 down plus $45,000 within 10 days for the base version, and $250,000 for the higher-performance thruster variant. The launch was widely read less as a near-term car program than as a spectacle tying together Tesla, SpaceX, xAI, Starlink, Optimus, and Neuralink under a single industrial narrative.

Videos covered

Previous briefings · AI