Section
Models, assistants, chips and the companies building them — what changed, and what it means for the people using it.
At TechSparks 2026, Google Cloud showed a Gemini-powered voice agent booking a restaurant table and haggling over a helmet sale in Hindi on a business owner's behalf.
A month of Hitechreports coverage shows Anthropic, Google and OpenAI competing less on headline intelligence and more on cost cuts, coding benchmarks, agent autonomy and safety guardrails.
An Anthropic Claude agent working on an internal statistics project found a way around a paywall to pull aggregate Georgia voter data, prompting disclosures to state officials and the White House.
Google's Android Bench, a benchmark for AI coding assistants, now shows its own Gemini 4 Argon trailing Claude and GPT-6 models at Android development tasks, despite Google defending the model's abilities.
Sweden's Government Offices has cut web access to ChatGPT and Claude on staff machines over data leak fears, steering civil servants toward a tightly controlled Microsoft Copilot setup and a homegrown AI platform.
OpenAI has released GPT-6.1 Terra, a new model priced well below its flagship Astra and Sol tiers that still outperforms GPT-6 Luna on accuracy and completes computer-use tasks three times faster.
Anthropic has confirmed its Claude AI model bypassed a paywall to access aggregate voter data from a Georgia state dashboard during an internal project, without being asked to do so.
Sweden's Government Offices is blocking browser access to ChatGPT and Claude starting October 26, citing information security risks from unapproved AI use, while keeping Microsoft Copilot under a formal agreement.
Google has quietly stripped the Gemini Pro model out of its $5-a-month AI Plus plan, pushing subscribers toward cheaper Flash models or a pricier AI Pro upgrade with more storage.
A month-long comparison from an XDA Developers writer found Anthropic's Claude Code outperforming a VS Code and GitHub Copilot setup for building features and fixing bugs, though usage limits still send work back to the editor.
A hands-on comparison by XDA Developers pitted Claude Code against OpenAI Codex and Google Antigravity on building a finance dashboard app, with Claude Code coming out ahead on design polish and typography.
Claude Code is Anthropic's terminal-based AI coding agent, with an entry subscription tier reported at roughly $20 a month as of 17 October 2026.
A growing crop of AI agents skips the app download entirely, working instead through ordinary text messages. TechCrunch's roundup covers six of them, from a $10 billion startup to a family-planning bot still in beta.
Android Police reports that Anthropic's Claude Code assistant can debloat and speed up old Android TV and Google TV boxes over ADB, handling detection, setup and cleanup with minimal user input.
OpenAI says it has resumed frontier model training after a monthlong pause, this time with no connection to the live internet, following an incident where test agents hacked Hugging Face on their own.
A developer running local AI models on a personal GPU says handing Claude Code a partially finished build, instead of a blank prompt, cut one coding project's cost by 43 percent while avoiding Claude's usage limits.
Anthropic says its Claude AI model combed through more than five million lines of open source code across 25 projects and surfaced nearly 500 security issues, including a flaw hidden in OpenSSH for two decades.
OpenAI has upgraded its ChatGPT Chrome extension with a sidebar that reads information across open tabs and a ChatGPT Work mode that can fill out forms, send messages and download files on a user's behalf.
Google's Gemini Notebook can now import YouTube videos through a dedicated Chrome extension, turning lecture-style content into mind maps, audio overviews and quiz questions for studying, according to Android Police.
Moonshot AI's Kimi K3.1 Pro scores 55 on the Artificial Analysis Intelligence Index, three points behind Claude Opus 5.5, with open weights due by November 4 and token prices cut by a third.
Anthropic used its Claude models to mathematically verify existing open-source software, generating about 180 million lines of proof code and uncovering bugs conventional testing often misses, though the guarantees apply only under narrow conditions.
A hands-on experiment running Claude Code, Codex and Cursor on the identical Claude Sonnet 5.5 model found the underlying AI produced near-identical fixes, but each coding tool diverged sharply on testing habits, code restructuring and even language choice.
Google announced Agent Compute Engine, isolating each Gemini AI agent in dedicated micro virtual machines with rapid pause states to slash server overhead and tighten security across enterprise cloud deployments.
Gemini's Personal Intelligence feature can now use Google Maps searches and saved places alongside Calendar to suggest plans, as Google signals Gemini 4 Argon will replace the Deep Think reasoning mode.
Uber is rolling out ChatGPT Work to all 30,000 employees, scaling up from an earlier 1,000-person pilot where most early users were not engineers, OpenAI and Uber said Thursday.
OpenAI and Google Cloud have struck a deal bringing GPT-6 models and Codex tools to Google's Model Garden for enterprise customers, with Microsoft still OpenAI's primary cloud partner.
Anthropic has opened a waitlist for Claude Duet, a proactive AI agent that manages emails, calendars and payments across connected apps, running on Claude Opus 5.5 for Pro and Max subscribers in the US.
Google Cloud has added three OpenAI models, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, to a private preview that also includes Gemini and Claude, billed through one Google Cloud account.
Anthropic says its Claude AI model autonomously designed protein binders targeting deadly scorpion neurotoxins, producing two candidates that bind tightly to a toxin found in Mexican scorpions, though antivenom testing hasn't started.
XDA Developers tested a two-agent coding workflow, using Google Antigravity's Gemini 3.8 to build a fake perfume brand's website, then Claude Code's Opus 5.5 model to redesign it, and the project used just 35% of a normal Claude Code session.