AI Token Price War: How DeepSeek Is Pressuring OpenAI and Anthropic
DeepSeek isn’t always the cheapest AI API: GPT-6 Luna and Claude Haiku 5.5 undercut its V4.1 Flash on token prices, but cache hits and a lean architecture cut its real costs.
DeepSeek isn’t always the cheapest AI API: GPT-6 Luna and Claude Haiku 5.5 undercut its V4.1 Flash on token prices, but cache hits and a lean architecture cut its real costs.
Meta’s Muse Wonder and Wonder Full add Active Memory and Active Focus. Both beat GPT-6 Astra and Claude Opus 5.5 on MemoryBench, and Muse Wonder costs $4/$20 per million tokens.
Meta’s Muse runs a cloud computer for every user, while GPT-6 Astra targets professional computer use. Agents that keep VMs running are pushing demand for server CPUs, not just GPUs.
Samsung hasn’t announced the Galaxy Fit4 yet, but an FCC filing for model SM-R391 shows it’s coming. Bluetooth 5.4 is confirmed; GPS, NFC and the release date are not.
The Galaxy Tab S12 Ultra looks just like the S11 Ultra. The changes are inside: a faster Dimensity 9500 and better cooling, from $1,399.99 in the US.
One UI 9 is rolling out to the Galaxy S26, S25 and S24. It adds Now Nudge, My FanCam, Security Brief and DeX upgrades; free at least 10GB before you install.
A 14-year-old PHP flaw found with AI lets attackers run code without logging in. Here are the fixed versions and the steps to secure your server.
Germany called the FSD name misleading, so Tesla renamed it in Europe. The system is still Level 2, and an EU-wide vote is expected in December.
GlobalFoundries will make silicon interposers for TSMC’s CoWoS packaging in New York. Interposers link GPUs and HBM in AI chips, and US-made supply starts ramping in 2028.
Hyundai’s Georgia plant builds robotaxis for Waymo and trains Boston Dynamics’ Atlas. Robotaxi deliveries start this quarter, while Atlas aims for plant work in 2028.