AI Agents
Google Cuts the Cost Curve for Agentic Work With Gemini 3.6 Flash
Google's Gemini 3.6 Flash launch turns token efficiency, latency and built-in computer use into a direct challenge for production agent budgets.
AI Models
AI Infrastructure
Microsoft Turns Mistral Into a European AI Capacity Partner
Microsoft's multibillion-dollar Mistral agreement makes European compute and sovereign deployment a product decision, not just a policy aspiration.
AI Infrastructure
Google’s reported 2028 chip plan puts inference efficiency at the centre of the AI race
A reported Google chip project would make tokens per watt, not raw model size, the next crucial benchmark.
AI Models
DeepSeek V4 Flash: pricing, capabilities and latest changes
DeepSeek V4 Flash is a low-cost one-million-token API model for automation, coding and agent workloads.
AI Models
Claude Opus 4.8: pricing, capabilities and latest changes
Claude Opus 4.8 is Anthropic’s current Opus-class model for demanding coding, agentic work and long-context analysis.
AI Models
Moonshot’s Kimi K3 Makes the Open-Model Race a Scale Contest
Chinese startup Moonshot AI has unveiled Kimi K3, a 2.8-trillion-parameter system it calls the world’s largest open-weight AI model. The company says the model performs competitively with leading US systems, according to Reuters’ July 17 report. The important qualifier is open-weight, not merely large. Moonshot is promising developers and companies a model they can inspect, adapt and run through their own infrastructure rather than rent solely through an American lab’s API. The reported scale
AI Models
Google puts a personal face on workplace AI video
Google is rolling out Gemini Omni video generation and personal avatars in Google Vids, turning a selfie and voice recording into a digital presenter for workplace videos. The company says the update, announced July 16, lets users generate or revise clips with text prompts and image references; Google's announcement says each AI-generated clip carries a digital watermark. The product sits inside Google Workspace's video tool, where the obvious use cases are training, internal announcements and
AI Models
Thinking Machines Opens Its First Model—and Makes Customisation the Bet
Thinking Machines Lab has released Inkling, its first public AI model, as open weights under an Apache 2.0 licence. The company’s model card says the July 15 release accepts text, image and audio inputs and produces text; Reuters reported the release as a potential alternative to open offerings from Chinese labs. Inkling is a 975-billion-parameter multimodal model, according to reporting and the company’s release materials. It is not being positioned as the top benchmark model: Thinking Machin
AI Models
Apple Clears a Regulatory Gate for iPhone AI in China
Apple Intelligence has been registered with China’s cyberspace regulator for use on iPhones, removing a major regulatory obstacle to the service’s long-delayed Chinese launch. Reuters reported the registration on July 15; Apple still has to turn approval into an actual rollout. The clearance matters because Apple has been selling AI-enabled iPhones elsewhere while its flagship generative features have been absent from one of its most important markets. Reporting from Reuters and other outlets
AI Models
Google faces fresh publisher challenge over Gemini training
Publishers and authors including Hachette, Cengage, Elsevier and Scott Turow have filed a class action accusing Google of using copyrighted books to train Gemini without permission. The TechCrunch reports on the new lawsuit was filed in the US District Court for the Southern District of New York and also alleges altered or removed copyright information. The plaintiffs say Google exceeded the limited permissions behind Google Books and Google Play by using copies for model training rather than
AI Infrastructure
Reflection locks in $1 billion of Nebius compute
Open-model startup Reflection AI has signed a $1 billion compute agreement with Nebius, giving it access to Nvidia's latest chips as it trains and serves models. TechCrunch reports that the deal follows a recent arrangement for SpaceX computing resources. Reflection, founded in 2024 by two former Google DeepMind researchers, is valued at $8 billion and has raised nearly $2.6 billion from backers including Nvidia, Sequoia and Lightspeed. Nebius is rapidly becoming a major supplier of scarce AI
AI Models
OpenAI’s GPT-5.6 Push Is Really a Pricing War for Enterprise AI
OpenAI rolled out GPT-5.6 on Thursday as a three-model family aimed squarely at enterprise buyers and coding teams, TechCrunch reports. The lineup splits into Sol, Terra and Luna, with OpenAI pitching Sol as its top coding system and saying it uses 54% fewer tokens on coding tasks than prior models. TechCrunch says OpenAI is also shipping a companion product called ChatGPT Work for document, spreadsheet and presentation tasks across desktop, web and mobile. The company is framing GPT-5.6 as it
AI Models
Ford Rehires 350 Veteran Engineers After AI Quality Checks Fall Short
Ford has rehired 350 experienced engineers — colloquially dubbed "gray beards" — after its AI-powered quality inspection systems fell short of expectations, according to Bloomberg. The veteran engineers, many of whom had been let go or retired, are being brought back to retrain staff, recalibrate automated tools, and address a surge in vehicle recalls that has pressured the automaker's quality ratings. Source: TechCrunch The move is a rare public admission from a major manufacturer that AI aut
AI Models
Mistral OCR 4 Turns Document Extraction Into Structured Enterprise Intelligence
Mistral AI has released OCR 4, a document intelligence model that represents a structural leap beyond traditional optical character recognition. Rather than outputting flat text, the model returns structured document representations complete with bounding boxes, block-type classification, and per-word confidence scores across 170 languages. Source: VentureBeat The model can be deployed as a single container on an organisation's own infrastructure — a capability Mistral is positioning directly
AI Agents
MIT and Microsoft's Murakkab Cuts AI Agent Energy Waste Without Sacrificing Performance
Source: MIT News — Researchers from MIT and Microsoft have developed Murakkab, an intelligent system that streamlines the design of AI agentic workflows and automatically optimises how those workflows are deployed in the cloud. The work will be presented at the USENIX Symposium on Operating Systems Design and Implementation. Agentic workflows chain together multiple AI models and external tools — databases, Python programs, APIs — to complete complex multi-step tasks like analysing video or pr
AI Models
OpenAI Unveils GPT-5.6 With Sol, Terra, and Luna — AI Gets a Pricing Ladder
OpenAI has begun a limited preview of GPT-5.6, splitting its next-generation model into three named tiers — Sol, Terra, and Luna — in a structural shift that introduces tiered pricing and two new reasoning modes, MarkTechPost reports. Sol is the flagship, targeting the highest-intelligence workloads with state-of-the-art performance on benchmarks like Terminal-Bench 2.1. Terra matches GPT-5.5 capability at roughly half the cost, positioned as the production workhorse. Luna is the low-cost opti