Fundació Escola Cristiana de Catalunya has introduced a six-level guide that makes the roles of people and artificial intelligence visible in schoolwork.
Moonshot has introduced Kimi K3, a 2.8-trillion-parameter model with native vision and up to one million tokens of context. Its headline figures are a reason to examine both capabilities and evaluation.
LM Studio has introduced Bionic, an initial-preview application for working with open models locally, remotely or through its cloud. The central idea is not only the agent, but the ability to choose where it runs.
ZKSecurity explains how its AI auditor found a vulnerability candidate in an OpenVM library. The issue was disclosed in May, fixed in version 1.6.0, and shows both the value and the limit of automated auditing.
AWS shows how Amazon Quick can prioritise opportunities, prepare meetings and update a CRM from connected data. Its usefulness will depend on permissions, source quality and review before it acts.
OpenAI introduces Work, a mode for research, finished deliverables, and persistent projects, with permissions and human review at its core.
Google DeepMind builds computer use directly into Gemini 3.5 Flash. The model can now see the screen, reason, and act across browsers, mobile apps, and desktops to automate enterprise tasks.
Google DeepMind unveils Gemma 4 12B, a multimodal model that processes image and audio without separate encoders and runs on laptops with 16 GB of memory.
OpenAI launches GPT-5.6 in three variants, touting efficiency, cybersecurity capabilities and aggressive pricing. The company directly compares its models to Anthropic's to claim the coding crown.
Anthropic's experiment managed to compile the Linux kernel with 16 Claude agents, but exposed the current limits of autonomous AI programming.
Google DeepMind launches an audio model that translates speech to speech in more than 70 languages while preserving the speaker's tone and rhythm. It's rolling out to Google Translate, Meet, and developers via the Live API.
AWS presents a reference architecture that uses Nova Act agents to run web flows, identify friction, and analyze UX-testing results.
Anthropic announced usage-based charges for its best model, but after several extensions the plan is on hold: subscribers keep 50% of their usage, with the additional 50% extended until August. Our analysis of the announcement, updated.
OpenAI trained GPT-5.2 and GPT-5.3 Codex entirely on NVIDIA infrastructure. The latter is its first agentic coding model to take part in its own development. Behind it lies a hardware race that now determines who gets to build frontier AI.
This website uses cookies to improve the browsing experience. Cookie policy.
No results. Try different words.
