OpenAI Launches Free AI Models That Run Outside the Cloud

OpenAI unveiled two open-weight language models, gpt-oss-120b and gpt-oss-20b. They are available for free download under Apache 2.0 and can run entirely on users’ own hardware.

The 20b version is small enough for a MacBook, while the 120b runs on a single GPU yet matches or exceeds OpenAI’s o4-mini on internal reasoning benchmarks. Both models share the same API and prompt format as OpenAI’s cloud models, enabling seamless swapping between local and hosted deployments.

Open-weight distribution lets enterprises meet strict privacy and security demands without leaving the OpenAI ecosystem. The strategy puts OpenAI inside laptops and private data centers, tightening customer loyalty even as deployment options diversify.

Every contributor Vivian Meng details how she is coaching her parents—and herself—to operate confidently in an AI-infused world. She shifts the focus from knowing how to use tools to mastering “epistemic fluency,” the ability to question and steer AI outputs.

Her essay recounts being stranded in cash-only situations in China and watching relatives rely effortlessly on superapps, illustrating how quickly technology can exclude. She offers concrete family exercises—trip planning, legal document summarizing, AI voice testing, and creative image generation—to build comfort and skepticism simultaneously.

Meng asserts that AI’s single, polished answers can erode personal agency unless users learn to interrogate them. She frames this literacy gap as a threat to dignity, warning that entire generations risk marginalization if they cannot discern what to trust.

SEO consultants and marketers are rebranding their craft as generative-engine optimization to secure visibility inside AI chatbots. The pivot follows sharp traffic declines from Google’s AI answers and new chatbot-style search experiences.

Many SEO teams have already been hit by layoffs, while startups such as Profound have raised tens of millions to sell analytics that track how models like ChatGPT and Gemini mention brands. Consultants now advise formatting content into citable chunks, listicles, and tables to increase the odds of being referenced in chat responses.

Chatbots trained on the entire web now deliver answers without sending users to source sites, severing the traffic lifeline that fueled a $75 billion SEO industry. With AI firms planning to sell their own ads, competition for what little referral traffic remains is already a bloodbath.

Scientist Krystal Maughan argues in The Guardian that the Global South is largely shut out of AI development and decision-making. This exclusion strips those nations of economic leverage and geopolitical voice in a technology shaping the future.

She cites repeated visa hurdles for African researchers at elite conferences like NeurIPS and the concentration of compute power in industrialised countries. The article also notes that most low-paid data-labelling work is performed in poorer regions while profits accrue in wealthy tech hubs.

Maughan warns that without data and labour sovereignty, AI risks repeating colonial trade patterns that kept the Global South dependent. She states that the technology’s long-term legitimacy hinges on including these perspectives.

Airbnb drops a host’s £12,000 damage claim after the guest proves the submitted photos were digitally manipulated. The company apologizes, refunds her £4,269 for the stay, and deletes the host’s negative review.

Initially, Airbnb sided with the “superhost” and ordered the guest to pay £5,314. The platform changed course five days after Guardian Money inquiries, warning the host for violating terms and launching a review of its investigation process.

AI image manipulation is now fueling false insurance and rental claims, as the article notes. This incident shows that major platforms struggle to authenticate digital evidence before penalizing users.