Anthropic and OpenAI Slash Prices in Hours as a CVSS 10 Bug Hits VeloCloud

AI & Machine Learning

xAI released Grok 4.7 on 21 September, its latest frontier model for coding, agentic tasks and professional knowledge work, built on a new base model with roughly 2.1 trillion parameters — a 40% expansion over Grok 4.6’s 1.5 trillion — following a longer reinforcement-learning run. The model works through demanding tasks for longer, reviews its own output more carefully, and ships with tighter safety calibration than earlier Grok releases, according to xAI. It carries a 500,000-token context window, accepts text and image input with text-only output, and is priced at $2 per million input tokens and $6 per million output tokens below a 200k-token prompt, rising to $4/$12 above that threshold. There is no waitlist: Grok 4.7 is live now in the Grok app, Cursor, Grok Build and the xAI API. The release lands after Elon Musk pushed the launch timeline back at least five times since late July, most recently saying on 11 September it “needs a few more days to cook.” Source: xAI Verified: True

Anthropic and OpenAI rolled out cheaper flagship models within hours of each other on 22 September, intensifying a price war despite both labs’ recent calls for an industry slowdown. Anthropic debuted Claude Opus 5.5, which it says matches the performance of its Fable 5.1 model while costing around 40% less to run than Opus 5, which shipped in July; Sonnet 5.5 and Haiku 5.5 are due “in the coming weeks.” OpenAI answered with two new GPT-6 tiers, GPT-6 Sol and GPT-6 Luna, cutting API prices by roughly 50% versus GPT-5.6’s promotional rates. The dueling releases came as both companies face growing competition from cheaper open-weight models out of China and elsewhere. Anthropic separately published research showing Claude can optimize more than 30 open-source biomolecular modeling tools for roughly 4x speedups, alongside a $1 million protein-design competition. Source: CNBC Verified: True

Consumer Hardware

Qualcomm opened its three-day Snapdragon Summit in Maui on 22 September by unveiling a dual-flagship strategy for the first time: the Snapdragon 8 Elite Gen 6 and a more powerful Snapdragon 8 Elite Extreme Gen 6. Both chips are built on a 2nm process around a new Oryon CPU design with two 5GHz prime cores and six 4GHz performance cores, and pair with the newly architected Qualcomm X105 5G modem for higher downlink and uplink speeds. Qualcomm says the standard Gen 6 delivers 85% better performance at 20% less power than last year’s Snapdragon 8 Elite, while the Extreme variant adds 50% more shared memory and an 85% faster NPU aimed at on-device agentic AI. Camera upgrades across both chips add Advanced Professional Video support, 8K60 capture and 4K240 slow motion. The chips will power the first wave of 2026 flagship Android phones expected later this year. Source: 9to5Google Verified: True

Peloton unveiled three new treadmills on 22 September, including its most affordable model yet, as the company leans further into running for a long-promised turnaround. The $2,195 Tread Flex is Peloton’s first foldable treadmill, cutting its footprint by close to half when stowed and topping out at 12.5 miles per hour with noise peaking around 70 decibels at full speed. It sits alongside an upgraded $3,495 Tread Vision with a new cushioning system and a $6,695 Tread+ Vision aimed at extreme-fitness users such as HYROX competitors. All three ship with new AI-powered coaching features and expanded fitness-tracker integration, including a new partnership with Whoop alongside existing Garmin, Apple Health and Fitbit support. The lineup goes on sale 1 October, backed by a “Let Yourself Run” ad campaign fronted by Miley Cyrus. Source: Peloton Interactive Verified: True

Cybersecurity

Arista Networks disclosed a maximum-severity flaw in its on-premises VeloCloud Orchestrator (VCO) on 22 September, warning that it is already being actively exploited. Tracked as CVE-2026-93952 and rated 10.0 on the CVSS 3.1 scale, the improper-input-validation bug lets a remote attacker with no login credentials reach privileged internal functions and potentially compromise the orchestrator and the SD-WAN Edge devices it manages. The flaw affects on-premises VCO running certificate-based Edge authentication — versions 5.2.3.15 and earlier, 6.1.3.7 and earlier, 6.4.2.7 and earlier, and 7.0.0.2 and earlier — while Arista’s hosted and dedicated VCO deployments were already patched. Fixed releases are out now for the 5.2 and 6.4 trains; patches for the 6.1 and 7.0 trains are still pending. Arista is urging on-prem customers to upgrade immediately or disable certificate-based authentication where a patch isn’t yet available. Source: Arista Networks Verified: True

E-commerce platform BigCommerce began notifying merchants on 17 September that attacker-compromised credentials for third-party Ribon apps were used to inject malicious scripts into online stores. BigCommerce says the credential compromise let attackers access shopper data across affected stores between 13 and 17 September before the access was cut off. The company has not yet disclosed how many merchants or customer records were affected, but it is urging store owners who used Ribon integrations to rotate API keys and review their storefronts for injected scripts. The incident adds to a string of e-commerce supply-chain breaches this year that have relied on compromised third-party app credentials rather than a direct platform vulnerability. Source: BleepingComputer Verified: True

Enterprise Infrastructure

Helsinki-based AI cloud provider Verda raised $189 million in a round led by Emergence Capital, announced 22 September, pushing its valuation above $1 billion and making it Europe’s newest AI infrastructure unicorn. Verda — rebranded from DataCrunch — builds and operates its own data centers in Finland running on 100% renewable Nordic power, offering on-demand GPU compute for AI training and inference without the procurement friction of the large hyperscale clouds. The company says it is generating a $165 million annualized revenue run rate and is already cash-flow positive, a rarity among AI infrastructure startups. Other investors in the round included MUFG Innovation Partners, Super Micro Computer and Varma Mutual Pension Insurance. The raise underscores growing investor appetite for regional “neoclouds” positioning themselves as leaner, greener alternatives to AWS, Azure and Google Cloud for AI workloads. Source: SiliconANGLE Verified: True

Policy & Regulation

New York Governor Kathy Hochul laid out implementation steps for the state’s RAISE Act on 21 September, and said her administration would explore requiring AI “kill switches” as a possible future safeguard. Under the RAISE Act, which takes effect 1 January 2027, large AI developers must register with the state starting in November, report critical safety incidents within 72 hours, and file quarterly assessments of catastrophic risk; Attorney General Letitia James can seek civil penalties up to $1 million for a first violation and $3 million for repeat violations. Hochul appointed tech attorney Marc Gilman as deputy director of the state’s new AI oversight office, DIGIT, to run implementation. She stressed the kill-switch idea remains preliminary — “all options are on the table,” with “no commitments at this time” — but said it would be considered alongside other safeguards in her next State of the State agenda. The move follows California’s recent AI safety legislation and comes as Congress remains gridlocked on federal AI rules ahead of the midterms. Source: amNewYork Verified: True